Banque de France WP #655 December 2017

Inflation Targeting as a Shock Absorber

Marcel Fratzscher1, Christoph Grosse Steffen2, & Malte Rieth3

December 2017, WP #655

ABSTRACT

We study the characteristics of inflation targeting as a shock absorber in response to large shocks in the form of natural disasters for a sample of 76 countries over the period 1970-2015. We find that inflation targeting improves macroeconomic performance following such shocks as it lowers inflation, raises output growth, and reduces inflation and growth variability compared to alternative monetary regimes. This performance is mostly due to a stronger response of monetary policy and fiscal policy under inflation targeting. Finally, we show that only hard but not soft targeting reaps the fruits: deeds, not words, matter for successful monetary stabilization.

Keywords: Monetary policy, Central banks, Monetary regimes, Dynamic effects

JEL classification: E42 ; E52 ; E58

1 DIW Berlin, Humboldt-University Berlin, CEPR, mfratzscher@diw.de 2 Banque de France, Monetary policy and financial research dept., Christoph.Grossesteffen@banque-france.fr 3 DIW Berlin, mrieth@diw.de Acknowledgements. We are thankful to Philippe Andrade, Agnès Bénassy-Quéré, Kerstin Bernoth, Fabio Canova, Michael Burda, Boris Hofmann, Athanasios Orphanides, Guillaume Plantin, Christian Proano, Federico Ravenna, Willi Semmler, Lutz Weinke and to participants of seminars and conferences at Computing in Economics and Finance New York USA 2017, International Association of Applied Economics Sapporo Japan 2017, Banque de France, DIW Berlin, and Humboldt University Berlin for useful comments and suggestions.

Working Papers reflect the opinions of the authors and do not necessarily express the views of the Banque de France. This document is available on publications.banque-france.fr/en

Banque de France WP #655 ii

NON-TECHNICAL SUMMARY

Inflation targeting has been praised for its success in bringing down inflation and raising credibility and accountability of policymakers. The costs of “cleaning up” after the global financial crisis, however, dramatically changed the perception of IT, since it may disregard the build-up of financial imbalances through its narrow focus on consumer price developments. As a result, scholars and policymakers today call for a refinement of the IT framework by allowing for more flexibility. This paper exploits the exogenous occurrence of natural disasters in order to answer the question whether countries operating under IT have a better macroeconomic performance in response to large adverse shocks than those with alternative regimes. We document important differences in the dynamic responses of countries under IT (targeters) and under alternative monetary regimes (non-inflation targeters) to large natural disaster shocks (Figure below).

Level effect of large natural disaster shocks in targeting and non-inflation targeting economies

Note: The figure shows the response of the level (cumulated first (log) difference) of key macroeconomic variables in both targeting (dark shaded area) and non-inflation targeting countries (light shaded area) to large natural disasters over the period 1970Q1-2015Q4. Statistical inference is based on 500 Monte-Carlo draws. Source: Fratzscher, Grosse Steffen and Rieth (2017)

Targeters perform significantly better regarding both the level and the volatility of output and prices. While GDP drops immediately under both regimes, the initial decline is smaller under IT and the subsequent recovery is stronger and faster. Four years after the shock, output is about two percentage points higher than under alternative regimes. Moreover,

Banque de France WP #655 iii

consumer prices increase significantly less for targeters. The difference to non-targeters is about six percentage points after four years. The results suggest that targeters rely on a different monetary-fiscal policy mix. The policy responses also significantly lower the shock-induced volatility of changes in output, consumer prices, consumption and investment relative to non-targeters. Lower volatility, in turn, is associated with smaller countercyclical private credit risk and lower term premia. The stability of consumer prices, in particular, translates into a more stable real exchange rate and implies relatively higher export and lower import growth under IT. Finally, we show that only hard but not soft targeting reaps the fruits: deeds, not words, matter for successful monetary stabilization. The paper provides an analysis of the mechanisms behind these results, as well as extensive robustness checks. The findings of the paper have a number of implications for central banks. They show that while IT may not strictly be a superior policy mandate in open economies in normal or tranquil times - as shown by the existing literature - it is a better mandate in crisis times, at least when the domestic economy is hit by a large real shock, such as a natural catastrophe. The better adjustment to large natural disasters of IT countries also suggests that IT has been more of a savior during the Global Financial Crisis than thought. However, this holds only if a central bank does not merely pretend to follow an IT strategy, but if it has gained credibility through a successful track record on IT. Therefore, it should be considered in the debate on reforming the present IT frameworks toward more flexibility that allowing for prolonged deviations from the target range come at a cost in terms of lower shock resilience.

Ciblage d’inflation comme absorbeur des chocs économiques

RÉSUMÉ Nous étudions les caractéristiques du ciblage de l'inflation en tant qu'amortisseur en réponse à de grands chocs sous forme de catastrophes naturelles pour un échantillon de 76 pays sur la période 1970-2015. Nous constatons que le ciblage de l'inflation améliore la performance macroéconomique à la suite de tels chocs car il réduit l'inflation, augmente la croissance de la production et réduit l'inflation et la variabilité de la croissance par rapport aux régimes monétaires alternatifs. Cette performance est principalement due à une réponse plus forte de la politique monétaire et de la politique budgétaire dans le cadre du ciblage de l'inflation. Enfin, nous montrons que seul le ciblage stricte mais pas le souple est fructueux: les actes, pas les mots, sont importants pour une stabilisation monétaire réussie. Mots-clés : Politique monétaire, Banques centrales, Régimes monétaires, Effets dynamiques Les Documents de travail reflètent les idées personnelles de leurs auteurs et n'expriment pas nécessairement

la position de la Banque de France. Ils sont disponibles sur publications.banque-france.fr

1

1. Introduction

Inflation targeting has become a dominant framework for monetary policy

over the last two decades. It has been praised for its success in bringing down

inflation and raising credibility and accountability of policymakers (Bernanke

and Mishkin, 1997; Ball, 2010). Its popularity has been reflected in an increasing

number of countries adopting inflation targeting (IT) at the end of 2015. The

global financial crisis has, however, dramatically changed the perception of IT as

an optimal framework for achieving macroeconomic stability, in particular at

times when the economy is confronted with large real or financial shocks. It has

been argued that IT, by focusing narrowly on inflation, may contribute to a

build-up of financial instability (Taylor, 2007; Svensson, 2010; Frankel, 2012),

lead central banks to neglect other important objectives, such as employment

(Stiglitz, 2008), and constrain monetary authorities in dealing with deep

recessions (Borio, 2014). As a result, scholars and policymakers call for a

refinement of the IT framework to allow for more flexibility (Svensson, 2009).

Many studies have analyzed whether inflation targeting affects the economic

performance of a country, though no clear-cut consensus has emerged in the

literature (Walsh, 2009; Ball, 2010). The focus of most papers in this literature

has been on the performance of IT regimes during the relatively good times of

the 1990s and till the beginning of the 2008 global financial crisis. During those

two decades of the “great moderation”, with declining and low inflation rates

amid strong economic growth, only few countries operating under an IT regime

experienced a deep economic or financial crisis. A different and arguably at least

equally important question is whether IT helps countries and their central

banks in dealing with crises, that is, whether it allows them to stabilize inflation

and output in response to large adverse shocks.

This paper focuses on this question and analyzes whether countries operating

under IT have a better macroeconomic performance in response to large

adverse shocks than those with non-IT regimes. Thereby, we empirically

respond to the question whether IT is a perpetrator, bystander or savior in the

wake of a crisis (Reichlin and Baldwin, 2013). We limit the analysis to the effects

of natural disasters, such as earthquakes or windstorms, as these are the most

2

exogenous large adverse shocks that can be identified and as they have been

shown to have a large impact on the macro-economy (Noy, 2009). Natural

disasters have a direct negative effect through the destruction of physical capital

and durable consumption goods such as housing and cars. The analysis can be

extended to include other types of shocks, such as financial shocks and other

real shocks, though this would entail the risk of endogeneity to the monetary

policy regime. At the same time share natural disaster shocks patterns with

shocks to the depreciation rate of capital or the quality of the capital stock that

have been used for the analysis of financial crises (Liu et al., 2011).

Natural disasters can be considered exogenous to the choice of the monetary

regime because they are largely unpredictable and not caused by economic

conditions. These features allow us to identify the conditional effects of IT using

relatively weak and verifiable assumptions about the distribution of the

unobserved factors that determine macroeconomic outcomes and about the

systematic relation between natural disasters and monetary regimes. In terms

of the treatment literature, we assume that conditioned on country

characteristics the “treatment” in form of a natural disaster is random, but

instead of focusing on the effects of the treatment, we are interested in whether

alternative monetary regimes imply different responses to the treatment.

To obtain a measure of such shocks, we use the reported insurance damage, in

percent of national GDP, from the EM-DAT dataset, which globally documents

natural disasters. As we are specifically interested in large real shocks, we focus

on disasters in the upper half of the disaster frequency distribution associated

with large declines in GDP. We match the shocks with quarterly macroeconomic

data for 76 countries over the period 1970Q1-2015Q4. We then estimate a set of

panel models similar to the single-equation regressions of Romer and Romer

(2004) and Kilian (2008), but extend this to a panel framework, to trace out the

responses of key macroeconomic variables to the shocks. The models contain

country fixed-effects as well as time-varying control variables that account for

alternative mechanisms which could impact and interact with the transmission

of natural disaster shocks.

We find that a natural disaster shock is contractionary and inflationary on

impact, followed by a short-lived boom in consumption and investment activity.

3

The empirical pattern resembles an adverse supply shock in a New Keynesian

model due to a destruction of physical capital and a decline in productivity,

followed by a subsequent positive demand shock (Smets and Wouters, 2007;

Keen and Pakko, 2011). The interpretation as an adverse supply shock is in line

with microeconomic evidence hinting toward economic disruptions that cause

indirect losses due to natural disasters. Inoue and Todo (2017) show that the

Great East Japan earthquake of 2011 was propagated via supply chain

disruptions to other regions. The substitution of production inputs poses a drag

on firm productivity. The subsequent investment boom can be understood

through the lens of the Solow (1956) growth model. A destruction of the

physical capital stock leads to a relative scarcity of capital for production which

induces catch-up growth through investment. Further, a destruction of durable

consumption goods likewise prompts a rise in private consumption.

We document important differences in the dynamic responses of countries

under IT (targeters) and under alternative monetary regimes (non-inflation

targeters) to large natural disaster shocks. Targeters perform significantly

better regarding both the level and the volatility of output and prices. While

GDP drops immediately under both regimes, the initial decline is smaller under

IT and the subsequent recovery is stronger and faster. Four years after the

shock, output is about two percentage points higher than under alternative

regimes. Moreover, consumer prices increase significantly less for targeters. The

difference to non-targeters is about six percentage points after four years.

These dynamics are reflected in statistically and economically significant

differences across monetary regimes in average GDP growth and inflation

following large shocks. The average quarterly growth rate of output is 0.2

percentage points higher under IT, and the average inflation rate is 0.4

percentage points lower. Moreover, there is robust evidence that inflation

targeters are more successful in stabilizing both output growth and inflation.

The standard deviation of both variables in the four years following a natural

disaster is only half of the size as under alternative monetary regimes.

While the main aim of the paper is to provide these stylized facts and we are

agnostic about the precise channels leading to the main results, we also provide

evidence on the potential mechanisms through which IT affects the adjustment

4

process to large shocks. The results suggest that targeters rely on a different

monetary-fiscal policy mix, they profit from reduced macroeconomic volatility

and lower financial frictions, and their external sector provides a better buffer.

Most strikingly, targeting central banks tighten monetary policy more or

loosen it less following the adverse shock to stabilize inflation, while fiscal

policy is accommodating. We interpret these findings in the light of a

coordinating role of IT for monetary and fiscal policy. A higher coefficient on

inflation in the loss function of central banks enforces a sound fiscal position

which, in turn, prevents from pro-cyclical public spending in the wake of large

disasters. Importantly, our regressions control for the institutional quality,

which are often held responsible for pro-cyclical fiscal policy in middle-income

countries (Frankel et al. 2013).

The policy responses also significantly lower the shock-induced volatility of

changes in output, consumer prices, consumption and investment relative to

non-targeters. Lower volatility, in turn, is associated with smaller

countercyclical private credit risk and lower term premia. The stability of

consumer prices, in particular, translates into a more stable real exchange rate

and implies relatively higher export and lower import growth under IT.

In contrast, non-targeting monetary authorities tend to ease policy more

aggressively and persistently in an effort to stabilize output, while fiscal policy is

contractionary. The adverse effects of private credit risk and term premia

reduce the effectiveness of monetary policy and this policy mix induces lower

output growth and higher consumer price inflation. Macroeconomic volatility is

also higher for non-targeters, and the strong increase in inflation leads to a

significant real appreciation of the currency.

Finally, we find that the better macroeconomic performance is only realized

by hard targeters that mostly comply with their inflation targets. There is only

limited evidence that countries which have introduced inflation targeting, but

deviate from their target for a prolonged period of time, reap the fruits of an

enhanced conditional macroeconomic performance. Here, compliance with an

inflation target is measured as the maximum period of consecutive recordings

of inflation rates outside of the target corridor in percent of periods since IT has

been introduced. This difference between hard and soft targeting is important

5

as it suggests that it is not the fact that a central bank formally adopts IT that

allows a superior performance. But this finding suggests that it is the track

record and the ensuing credibility of an IT central bank that allows it as well as

the fiscal authorities to respond differently and more successfully to the

economic shock of a natural disaster.

Our paper relates to a large literature on the impact of IT on macroeconomic

outcomes. In a seminal contribution, Ball and Sheridan (2004) find no

significant differences between IT and non-IT countries, as measured by the

behavior of inflation, output, and interest rates, in a sample of OECD member

states and based on a difference-in-difference approach that controls for

regression to the mean. Similarly, Lin and Ye (2007) detect no effect of IT on

either inflation or inflation variability in industrial countries when employing

propensity score matching methods. Using OLS to study the impact of IT on

disinflation periods in OECD countries, Brito (2010) concludes that inflation

targeters were not able to bring inflation down at less output costs than non-

targeters. On the other hand, Gonçalves and Carvalho (2009) find in a sample of

OECD countries that inflation targeters suffer significantly smaller output losses

for bringing down inflation when controlling for possible selection bias through

Probit or Heckman regressions. Moreover, Lin and Ye (2009) and Lin (2010)

show evidence based on propensity score matching that IT does lower inflation

and inflation variability in developing countries.

These apparently contradicting findings extend to studies which focus on the

performance of IT during specific time periods and across different country

samples. Concerning the global financial crisis, Rose (2014) finds that IT did not

substantially change how a country weathered the crisis. By contrast, Carvalho

Filho (2010) and Andersen et al. (2015) present evidence that IT countries fared

significantly better than others during this episode. Regarding different country

samples, De Mendonça and e Souza (2012) find, based on propensity score

matching, that IT may be particularly beneficial in developing countries,

suggesting that IT might work if it helps improve the credibility of monetary

authorities.

To the best of our knowledge, our paper is the first one to analyze whether

inflation targeting is effective as a shock absorber in response to large real

6

shocks. Our econometric approach, which has not been used before to study the

macroeconomic impact of IT, has several advantages. First and foremost,

estimating the conditional effect of IT, given an exogenous event, bypasses the

need to directly deal with the potentially endogenous choice of the monetary

regime to macroeconomic conditions as it “nets out” the unconditional impact of

IT on the response variables. The methodological approach is inspired by

Ramcharan (2007), who uses disasters to evaluate the effects of alternative

exchange rate regimes on the adjustment to real shocks in an annual sample of

developing countries and employing pooled OLS.

The remainder of the paper is structured as follows. Section 2 formulates the

main hypotheses. It also describes the empirical strategy and the data. Section 3

contains the core results. Section 4 provides a sensitivity analysis, before the

final section concludes.

2. Empiricalstrategyanddata

We characterize in this section potential channels for how inflation targeting

affects the policy response to and propagation of large natural disaster shocks in

an economy. This reasoning is then used to derive the empirical hypotheses. We

then describe the dataset, the data transformation and the empirical model used

to test our main hypotheses.

2.1. Inflation targeting and effects of large natural disaster shocks

We consider, in line with the literature, that natural disasters correspond to

an adverse shock to physical capital and durable consumption goods.1 The

empirical response to such a shock, which we intend to trace out in this paper,

will be affected by two factors, first the policy response to the shock, and

second, the propagation of the shock within the economy. We suppose that the

policy response and the propagation of the shock are affected by the choice of

the monetary policy regime.

Inflation targeting (IT) might affect the policy response to a natural disaster

shock along two dimensions. First, it imposes constraints on policymakers.

1 These shocks share essential features with shocks to the quality of capital or the capital depreciation rate, which have been at the heart of the global financial crisis. One important caveat applies to this generalization. Liu et al. (2011) show in an estimated DSGE model of the US economy that while a shock to the rate of capital depreciation is contracting output it is also disinflationary. In contrast, our natural disaster shock leads to a rise in inflation on impact in the data, as we show below.

7

Following Svensson (2010), IT can be described as a monetary framework

under which the central bank publicly announces an official numerical target or

target range for the inflation rate over a specific time horizon. Monetary policy

under IT is therefore often described as ‘constrained discretion’ (Bernanke and

Mishkin, 1997; Kim, 2011). IT is typically associated with enhanced

communication standards of monetary authorities with the public and aims at

increasing accountability, possibly through implicit incentives or explicit

contracts for central bankers. The monetary authority also explicitly

communicates to the public that low and stable inflation is its main goal, bases

its decisions on inflation forecasts, and enjoys a high degree of political

independence.

Second, the aforementioned features in the conduct of monetary policy under

IT ideally make announced inflation targets of the central bank more credible.

The main advantage over alternative monetary regimes is thus that IT

addresses the dynamic consistency problem (Kydland and Prescott, 1977).

Clarida et al. (1999) show that stronger commitment allows the central bank to

affect agents’ inflation expectations directly. Lower and better anchored

inflation expectations, in turn, reduce the short-run tradeoff between inflation

and output (Walsh, 2009). According to a forward looking Phillips curve,

inflation depends on future output gaps. A natural disaster shock is lowering

potential output through the destruction of productive capital. All else equal,

lower potential growth closes the output gap, which ultimately raises inflation.

The central bank would like to give the signal that it will be tough in the future

without contracting demand much today. This strategy would lower inflation

today, while keeping output at potential. However, such a strategy is only

credible under commitment, which IT facilitates to attain.

Bernanke and Mishkin (1997) highlight that lower uncertainty about future

inflation supports savings and investment decisions and reduce the riskiness of

nominal financial and wage contracts. This can impact the propagation of

natural disaster shocks for two reasons. First, lower nominal uncertainty in

wage contracts might allow for higher employment in response to natural

disaster shocks. Strulik and Trimborn (2014) show in a macro model that the

GDP impact of natural disasters is affected by the households’ labor response. In

8

their model, a natural disaster shock destroys physical capital and durable

consumption goods, such as residential housing. Households want to provide

more labor in response to a natural disaster in order to rebuild housing, which

enters their utility function directly and exhibits a high relative marginal utility.

This response in labor supply partially off-sets the negative effect on GDP due to

the destruction of physical capital. This off-setting effect is stronger if firms are

more willing to demand labor, which is more likely if they face less nominal

wage uncertainty. This makes the drop in GDP on impact possibly less severe

under IT, while simultaneously leading to a rise in consumption activity related

to durable goods.

Second, investment activities in a reconstruction-led Keynesian boom can be

positively affected by IT through lower riskiness in nominal credit contracts and

higher savings (Benson and Clay, 2004). While this has a dampening effect on

the short-run decline in GDP in response to a natural disaster, the literature is

inconclusive whether there is a medium to long-term positive growth effect

(Skidmore and Toya, 2002). Hallegatte and Dumas (2009) investigate in how far

the long-run effects of natural disaster shocks depend on the quality of

technologies that replace the damaged productive capital. They find in a model-

based analysis that a higher level of embodied technological change after

reconstruction following a natural disaster can increase the level of production,

but does not affect the long-term growth rate. Finally, the response in

investment to natural disasters also depends on countries’ capacity to fund the

reconstruction. Inflation targeting might lower credit constraints through

higher savings and lower nominal uncertainty, therefore being conducive for a

reconstruction-led boom and eventually promoting a faster embodiment of

technological change.

These considerations lead us to the following hypotheses. When a country is

confronted with large natural disaster shocks, we first expect that inflation

targeting lowers the response in inflation and cushions the drop in output

growth, and second, that inflation targeting reduces the variability of both

inflation and output growth. In the following we present the data as well as the

empirical strategy to test these two hypotheses.

9

2.2. Data description

2.2.1. Large Natural Disaster Shocks.

We use the EM-DAT database from the Center for Research on the

Epidemiology of Disasters (CRED) to select large natural disasters.2 The

database provides detailed information on disasters such as earthquakes, floods

and storms, among others, which occurred worldwide since 1900. The data is

compiled from various sources, including UN agencies, non-governmental

organizations, insurance companies, research institutes, and press agencies.

There are low threshold criteria for events to be registered. One condition out of

the following four needs to be met: 10 or more people are killed; 100 or more

people are affected; there is a declaration of a state of emergency; there is a call

for international assistance. The low thresholds lead to many observations that

require filtering for large disasters, as our research question focuses on extreme

shocks which have economic consequences on a national scale.

To this end, we follow the existing literature on the macroeconomic

consequences of disasters (Noy, 2009). In particular, we use the reported

estimated damage, which is the direct damage to property, crops, and livestock,

reported in US dollars and valued at the moment of the event. To be as precise

as possible, we weight the reported damage according to the occurrence of the

event within a quarter, reflecting that a natural disaster taking place at the

beginning of the quarter has a larger impact on quarterly measures of

macroeconomic variables than one towards the end. The weighted reported

damage is calculated as wDAM = DAM(3-OM)/3, where OM denotes the onset

month, that is, the reported starting month of the natural disaster. In the

sensitivity analysis we show that our results are robust to alternative weighting

schemes. Next, we sum over all weighted damages across events within the

same quarter that are classified as natural disasters.3 This is motivated by our

research question, which focuses on the consequences of extreme shocks on the

economy in general, abstracting from the specific type of event that lead to the

2 Guha-Sapir, Below, Hoyois – EM-DAT: International Disaster Database – www.emdat.be – Université Catholique de Louvain, Brussels, Belgium.

3 These fall in either of the following categories: geophysical, meteorological, hydrological, climatological, biological and extraterrestrial. We exclude technological disasters.

10

damage. We standardize the disaster size by dividing the weighted and

aggregated reported damage by the level of nominal GDP one year prior to the

event.

The selection of natural disasters leaves us with 1.953 events over the years

1970 to 2015. We further reduce the number of events in two steps as we are

interested in the economic adjustment process to real shocks that are of

economic relevance to a country and since we aim at eliminating noise in the

reporting of disasters. First, we take the upper 50th percentile of the

constructed damage variable, that is, those with relatively large direct costs.

Second, we select from those the episodes which are associated with large drops

in GDP growth. We do this by computing the percentiles of contemporaneous

GDP growth relative to trend conditional on a large disaster to occur and select

the bottom 50th percentile. This procedure gives us 254 events that we use as

shocks for the subsequent analysis: 79 large disasters under IT and 175 shocks

under alternative monetary regimes.

Figure 1: Distribution of large disaster shocks in targeting and non-inflation targeting countries over time

Note: The figure shows the mean shock in inflation targeting (as of 2013Q1) and non-inflation targeting countries as percent of GDP(t-4) over time and the average number of large natural disasters in both country groups over time.

0.1

.2.3

Mea

n sh

ock

acro

ss c

oun

trie

s in

% o

f GD

P(t

-4)

1970q3 1985q3 2000q3 2015q3

Inflation targeting countries

0.1

.2.3

1970q3 1985q3 2000q3 2015q3

Nontargeting countries

0.0

5.1

.15

Ave

rage

num

ber

of s

hock

s

1970q3 1985q3 2000q3 2015q3

0.0

5.1

.15

1970q3 1985q3 2000q3 2015q3

11

We focus on those disaster episodes that lead to a substantial reduction in GDP

growth since we are not interested in the macroeconomic effects of natural

disasters per se but in an evaluation of the conditional impact of alternative

monetary regimes given a large shock. The reported insurance damage in US

dollars is only a measure of the direct macroeconomic effects of natural

disasters. The selection of large reductions in GDP growth allows us to capture

also the indirect effects, such as supply chain disruptions as documented by

Inoue and Todo (2017) that might vary across disasters. The selection of large

contemporaneous drops in GDP growth is also backed by recent findings in the

literature on the growths effects of natural disasters (Noy, 2009; Felbermayr

and Gröschl, 2014). Moreover, in the sensitivity analysis we show that our main

results hold when using different thresholds in the first step of the shock

selection procedure or when leaving out the second step.

2.2.2. Inflation targeting.

Regarding the monetary regime, we distinguish between inflation targeting

and non-inflation targeting regimes. The dates at which a country adopted IT

feature some heterogeneity in the literature, depending on the criteria used.

While some studies classify a monetary authority to follow IT after simply

having announced numerical targets for inflation, others use dates when IT has

been effectively implemented. This implementation implies that other nominal

anchors like exchange rate targets are abandoned.4 For our analysis, we follow

Roger (2009) and create a dummy variable for the quarter-country pairs with

an effectively implemented IT regime.5 We consider the European Central Bank

to follow an implicit IT regime and declare countries that have adopted the euro

as inflation targeters when they enter the common currency. Member countries

which introduced IT before joining the euro are classified as targeters from the

initial adoption of IT onwards. In a sensitivity analysis we show that the results

are not driven by this classification of euro area countries. Table A1 in the

4 The difference between these two dating conventions is referred to in the literature as ‘soft IT’ versus ‘fully fledged IT’ (Vega and Winkelried, 2005).

5 We update this list with countries that have adopted inflation targeting since 2007 by collecting data available from central bank websites.

12

Online Appendix provides an overview of IT and euro adoption dates. In the

sample, we have 41 countries that adopted IT and 35 countries that did not.

Figure 1 brings together the data on natural disasters and inflation targeting.

It shows the distribution of the mean size and the average number of large

disasters for targeters and non-inflation targeters over time. While the average

size and frequency of the shocks tend to be larger for the former, both groups

are affected by disasters. Importantly, in the group of countries that adopted IT,

there are large and numerous shocks both before and after the spreading of this

monetary regime in the 1990s and 2000s. Finally, the figure indicates an overall

increase in the number and size of disasters across time. This is a well-known

fact in the literature on natural disasters, which we aim to capture through time

fixed effects in the empirical implementation.

2.2.3. Other macroeconomic data and controls.

We collect macroeconomic data at a quarterly frequency for the period

1970Q1 to 2015Q4. The cross-section contains 76 countries, mostly advanced

economies and emerging markets. The panel dimensions are dictated by the

joint availability of the main variables used in the analysis. We start 20 years

before the first introduction of IT in New Zealand in 1990Q1 to increase the

estimation precision for the control group of countries without IT. The results

are insensitive to using only observations from 1990 onwards. In any case, one

has to bear in mind when interpreting the results that the sample, even though

it contains the global financial crisis, is likely to be influenced by the period of

the “great moderation”.

Table A2 in the Online Appendix lists all countries in the sample. We obtain

real and seasonally adjusted data on output, private and government

consumption, investment, exports and imports from the OECD national accounts

statistics, as well as from national sources. If seasonal adjusted data are not

available, we make this transformation by our own. We further obtain CPI price

indices, money market rates, and lending rates from the IMF International

Financial Statistics. We compute CPI-based real exchange rates relative to the

US using bilateral nominal exchange rates and CPI differences as real effective

exchange rates are not available at the quarterly frequency for our country

13

sample. The policy rate of the monetary authority and the three month T-Bill

rate are from Datastream, while bank lending rates and longer term

government bond yields are taken from the IMF International Financial

Statistics database.

We clean the data with respect to periods of extraordinary large nominal

fluctuations. Specifically, we drop observations for all variables during periods

of extremely high nominal volatility, when either the policy, the inflation or the

nominal exchange rate exceeds a given threshold of quarterly rate of change. We

set relatively high thresholds with the aim at only eliminating periods of large

volatility that are due to hyperinflations and not the result of a large shock from

a natural disaster, or the global financial crisis. After dropping these periods, we

country-wise also drop observations that are separated along the time

dimension from the longest continuous sequence of observations to ensure that

the country time-series are uninterrupted. We thereby mostly eliminate periods

in the 1980s and 1990s. Only for six countries we drop data spanning the global

financial crisis and these episodes are driven by extraordinary country-specific

events, like in the Ukraine. Moreover, we have verified that our results are

largely insensitive to alternative thresholds.6

Finally, we collect a number of control variables. We obtain annual data on

total population and the degree of urbanization as a percentage of the total

population living in cities from the World Bank. These country characteristics

have been used in the literature to control for the possible differential impact a

natural disaster may have across countries given regional differences. As a

proxy for the level of democracy, we use the polity IV variable from the Center

for Systemic Peace. Table A3 in the Online Appendix provides an overview of

the variables and sources.

2.3. Empirical model and identification

In a first specification, we measure the average dynamic effect of our shock

variable on the macro-economy across countries, using the following model:

6 The main results are based on thresholds of 20 percent for the quarterly change of inflation, 40 percent for the nominal exchange rate, and 20 percent for the policy rate. They hold when changing the thresholds by ±10 percentage points.

14

��,� = � + � + � + � ���,���

�

���+ �����,���

�

���+ ���,��� + ��,� . (1)

��,� denotes the quarterly rate of change in the dependent variable for country

i in quarter t. The main endogenous variables of interest are changes in GDP,

consumer prices, and the policy rate. The natural disaster shock is captured by

��,��� .

To account for time-invariant country characteristics, such as the geographic

exposure to large natural disasters, we include country fixed effects�.

Moreover, we allow for year fixed effects� to correct for common

unobservable time-varying factors, such as global growth and inflation trends as

well as climatic change. To remove possible autocorrelation in the error term,

we include lags of the dependent variable. This makes our approach similar to

the single-equation regressions of Romer and Romer (2004) and Kilian (2008)

for the analysis of monetary policy and oil supply shocks, respectively. In the

sensitivity analysis, we use alternative estimators to confirm that our results are

unlikely to suffer from the Nickell bias (Nickell, 1981), as can be expected in our

sample where the time-dimension is long (T>30, see Judson and Owen, 1999).

Finally, we add several time-varying control variables in the vector��,���. They

include the degree of urbanization, population density and a measure for the

level of democracy and enter at a lag of four quarters in order to prevent

endogenous feedback in response to the natural disaster shock. We set J = 15

and L = 4 to obtain impulse responses over a horizon of four years and to ensure

that the residuals are free of autocorrelation. We estimate (1) by OLS based on a

within-transformation, assuming that the error term ��,� is independent and

identically distributed���(0, "#$).

In a second specification, we extend the framework by including the IT-

dummy, denoted by�&�,���, in levels and as an interaction term with the shock to

estimate the differential effect of inflation targeting in the aftermath of large

natural disaster shocks:

��,� = � + � + � + � ���,���

�

���+ �'����,��� + (��&�,�����,��� + )��&�,���*

�

���+ ���,��� + ��,� (2)

The main parameters of interest are now the (�′,. They capture the difference

between the dynamic effects of large real shocks under inflation targeting and

15

non-inflation targeting regimes. The specification thereby also relaxes the

standard assumption in panel data models of common slopes across all panel

units. Throughout, we base statistical inference on 500 Monte Carlo draws.7

To illustrate the identification strategy, we consider the case of J = L = 0 and

summarize all explanatory variables in (2) outside the brackets in the vector -�,�.

Further, we define as E/��,�0��,� > 0, -�,�2 the expected value of ��,� given that

a natural disaster occurred and conditioned on the set of co-variates -�,�. Following Ramcharan (2007), the average effect of the disaster is then

We make two assumptions to simplify (3). First, the residual ��,�, which

captures unobserved drivers of ��,�, is unrelated to the occurrence of the

disaster shock ��,�. The assumption is motivated by the random nature of these

shocks and our strategy of accounting for country characteristics that capture

the general susceptibility to these shocks. Then,

E/��,�0��,� > 0, -�,�2 = E/��,�0��,� = 0, -�,�2 = 0. Second, natural disaster shocks do not systematically affect the choice of the

monetary regime. This assumption is motivated, on the one hand, by the

remarkable stability of inflation targeting as a monetary regime (Rose, 2007;

2014). No country that adopted IT has ever abandoned it. This stability rules out

the possibility that a country abolished IT in response to a large natural

disaster. On the other hand, it is easy to check whether in our sample countries

adopted IT (in the four years) following a large shock. We find only three such

cases and excluding these countries from the analysis does not change the

results. Based on these descriptive statistics we can essentially exclude the

possibility that the decision to inflation target depends on disaster realizations.

Nevertheless, we also test this assumption formally by estimating a set of probit

models where the dependent variable is the probability that a country targets

7 Following Romer and Romer (2004), we use the estimated covariance matrix of the coefficients to draw new coefficients

from a multivariate normal distribution, from which we compute a distribution of impulse responses.

E/Δ��,�0��,� > 0, -�,�2 − E/Δ��,�0��,� = 0, -�,�2 = ����,� + (�E/�&�,�0��,� > 0, -�,�2��,�

+)�'E/�&�,�0��,� > 0, -�,�2 − E/�&�,�0��,� = 0, -�,�2* (3)

+'E/��,�0��,� > 0, -�,�2 − E/��,�0��,� = 0, -�,�2*

16

inflation. We include several institutional and economic country characteristics

as explanatory variables, following Lin and Ye (2007), Gonçalves and Carvalho

(2009), and Lin (2010), in addition to geographic factors measuring the

exposure of countries to natural disasters. Moreover, we add our shock variable

with lag 0-15 to test whether it affects the probability of targeting. Table A4 in

the Online Appendix shows that none of the shock lags is individually

significant, and that they are jointly insignificant as well. The p-values of the

corresponding F-tests are all close to one, depending on the specification. All in

all, we hence assume that E/�&�,�0��,� > 0, -�,�2 = E/�&�,�0��,� = 0, -�,�2 = �&�,�. Under

these two assumptions (3) simplifies to

E/Δ��,�0��,� > 0, -�,�2 − E/Δ��,�0��,� = 0, -�,�2 = ����,� + (��&�,���,� ,

where (� measures the difference between the average effect of the shock in

targeting and non-inflation targeting regimes. However, to attach a causal

interpretation to (�, we need to carefully control for other potential country

features that could affect both the choice of the monetary regime and the

response of the economy to the shock. In the sensitivity analysis, we will

therefore in particular control for the level of development of a country, as there

is some evidence that the introduction of IT is mostly relevant for the

performance of developing countries (Lin and Ye, 2007; Ball, 2010; Lin, 2010),

and for the exchange rate regime (Ramcharan, 2007), as well as for other

variables which have been used in the literature on natural disasters to account

for different shock absorption capacities of countries.8

3. Inflationtargetingandmacroeconomicperformance

In this section, we first present the estimated average dynamic effects of large

real shocks on the macro-economy. Then we test whether IT economies

respond differently to these shocks from non-inflation targeting economies.

Finally, we highlight several channels through which IT may alter performance.

8 An alternative identification strategy that eliminates the second step in the shock selection procedure (see Section 2.2.1) would be to use the natural disasters as an instrument for GDP growth and then assess the differential effects of IT given an exogenous change in GDP growth. This approach is not ideal for our research question, however, as we are interested in the response of GDP growth (and volatility) itself under IT.

17

Figure 2: Dynamic effects of large natural disasters on the first (log) differences of key variables

Note: The figure shows the average response of the first (log) differences of key macroeconomic variables to large natural disasters in a sample of 76 inflation targeting and non-inflation targeting countries over the period 1970Q1-2015Q4. Statistical inference is based on 500 Monte-Carlo draws.

3.1. The dynamic effects of large natural disasters

Figure 2 summarizes the point estimates of the dynamic effects of large

disasters on the change in key macroeconomic variables on average across

monetary regimes, and their 68 and 90 percent confidence bands. We find a

significantly negative effect on GDP growth, which declines by about 0.3

percentage points upon impact. It then overshoots for roughly two quarters,

before returning to the level where it would have been without the shock.9

The economic consequences of natural disasters can be viewed as those of a

negative shock to the capital stock of an economy which distorts production.

They typically cause direct damages to houses and contents, machinery, and

infrastructure as well as indirect impacts due to business interruption. Post-

disaster, the replacement of destroyed capital through more productive

9 Our analysis does not aim at contributing specifically to the literature on the growth effects of natural disasters, which has not come to a consensus. Cavallo et al. (2013) find no significant effect of large natural disasters on GDP growth once controlling for political turmoil occurring in the aftermath of natural disasters. Loayza et al. (2012) find negative growth effects only for a subset of natural disasters, like earthquakes, windstorms, and droughts, while floods tend to have a mildly positive impact. Kousky (2014) provides a survey of this literature.

-.4

-.2

0.2

.4P

erce

nt

0 2 4 6 8Quarters

D.GDP

-.2

0.2

.4.6

Per

cent

0 2 4 6 8Quarters

D.Prices

-.2

-.1

0.1

.2P

erce

nt

0 2 4 6 8Quarters

D.Policy_rate-.

50

.5P

erc

ent

0 2 4 6 8Quarters

D.Gov_consumption

-.2

0.2

.4P

erc

ent

0 2 4 6 8Quarters

D.Priv_consumption

-1-.

50

.51

Pe

rcen

t

0 2 4 6 8Quarters

D.Investment

-2-1

01

2P

erce

nt

0 2 4 6 8Quarters

D.RER

-1-.

50

.51

Per

cent

0 2 4 6 8Quarters

D.Exports

-1-.

50

.51

Per

cent

0 2 4 6 8Quarters

D.Imports

18

investments and new technologies, spending of insurance payouts, and possible

multiplier effects of increased household and business outlays generate catch-

up demand and increase GDP growth.

As production is interrupted, various products - and labor - are in short

supply, and more expensive substitutes are used, inflation increases

significantly; by about 0.2 percentage points upon impact. When demand surges

in the following quarters, inflation rises further to roughly 0.4 percentage points

above trend, before the effects fade out. Despite the immediate price pressure,

central banks on average aim at countering the drop in output growth by

lowering policy rates. When GDP growth recovers and inflation rises further,

there is a tendency to tighten policy.

Looking at the components of domestic absorption, government consumption

seems largely unresponsive, leaving private consumption and investment as the

main drivers of the overshooting of output growth. They both rise significantly

in the quarter following the shock. Regarding the external sector, the real

exchange rate is not affected much upon impact, as higher inflation and a

weaker currency balance each other out, on average – with a depreciation

contributing to an increase in inflation via higher import prices – but it then

significantly appreciates when inflation reaches its peak and policy rates are

raised.10 Real export growth drops immediately after the shock, contributing to

the initial decline in GDP growth, before recovering. Import growth, on the other

hand, tends to be above trend simultaneously with the other domestic demand

components, and in line with the appreciated real exchange rate. Overall, these

findings on the macroeconomic effects of natural disasters are by and large in

line with the literature (Noy, 2009; Felbermayr and Gröschl, 2014).

Figure 3 shows the dynamic effects of large natural disasters in cumulative

terms. The shocks have long-lasting and significant effects on GDP, several of its

components, and consumer prices. After the negative impact of the natural

disasters, the affected economies start recovering and return to the pre-crisis

level of GDP after one to two years. Net exports are a drag on GDP as exports

persistently decline and imports increase, consistent with the propensity of the

10 A decline in the exchange rate implies an appreciation of the currency.

19

currency to appreciate in real terms. The appreciation in turn appears to be a

result of the sustained increase in the domestic price level. Finally, monetary

and fiscal policy appear largely neutral over this horizon. These average effects,

however, mask important differences in the conduct of both monetary and fiscal

policy across monetary regimes, as we will show next. This in turn has

significant implications also for the differential evolution of the other variables.

Figure 3: Dynamic effects of large natural disasters on the level of key macroeconomic variables

Note: The figure shows the average response of the level (cumulated first (log) difference) of key macroeconomic variables to large natural disasters in a sample of 76 inflation targeting and non-inflation targeting countries over the period 1970Q1-2015Q4. Statistical inference is based on 500 Monte-Carlo draws.

3.2. The effects of inflation targeting on macroeconomic dynamics

We now assess whether and how inflation targeting changes the dynamic

adjustment to large real shocks by computing impulse responses for targeting

and non-targeting economies. We then test whether the responses are

significantly different across regimes. For short, we refer to countries operating

under IT as targeters and to economies with non-IT regimes as non-targeters,

although technically we are using only the within variation in the data given that

our model contains country fixed effects. We concentrate on the level effects,

-.4

-.2

0.2

.4.6

Per

cent

0 2 4 6 8Quarters

GDP

01

23

Per

cent

0 2 4 6 8Quarters

Prices

-.4

-.2

0.2

.4P

erce

nt

0 2 4 6 8Quarters

Policy_rate

-1-.

50

.51

Pe

rcen

t

0 2 4 6 8Quarters

Gov_consumption

-.5

0.5

1P

erc

ent

0 2 4 6 8Quarters

Priv_consumption-1

01

23

Pe

rcen

t

0 2 4 6 8Quarters

Investment

-3-2

-10

12

Per

cent

0 2 4 6 8Quarters

RER

-2-1

01

Per

cent

0 2 4 6 8Quarters

Exports

-10

12

3P

erce

nt

0 2 4 6 8Quarters

Imports

20

which provide a clearer picture, and defer the underlying responses of first (log)

differences to Figure A1 and Figure A2 in the Online Appendix.

Figure 4: Level effects of large real shocks in targeting and non-inflation targeting economies

Note: The figure shows the response of the level (cumulated first (log) difference) of key macroeconomic variables in both targeting (dark shaded area) and non-inflation targeting countries (light shaded area) to large natural disasters over the period 1970Q1-2015Q4. Statistical inference is based on 500 Monte-Carlo draws.

Figure 4 shows the adjustment of both groups to the shock. There are a

number of significant differences. First and foremost, output is significantly

higher and prices increase significantly less under IT. In fact, output persistently

rises above the level prevailing in absence of the shock, whereas it falls below

the pre-shock level for non-targeters. Consumer prices tend to rise under both

regimes, but only mildly and mostly indistinguishable from zero under IT, while

there is a strong and long-lasting price increase otherwise. Together, these

findings provide preliminary support for our first hypothesis.

Several mechanisms are relevant for understanding the pronounced

differences. Regarding the response of economic policy, targeters tentatively

rely on fiscal policy to buffer the adverse shock, whereas non-targeters strongly

use monetary policy for that purpose. Specifically, in the latter group central

banks aggressively lower policy rates, by cumulatively two percentage points

-3-2

-10

12

Per

cent

0 2 4 6 8 10 12 14Quarters

GDP

05

10

15

Per

cent

0 2 4 6 8 10 12 14Quarters

Prices

-3-2

-10

1P

erce

nt

0 2 4 6 8 10 12 14Quarters

Policy_rate

-6-4

-20

2P

erce

nt

0 2 4 6 8 10 12 14Quarters

Gov_consumption

-4-2

02

4P

erce

nt

0 2 4 6 8 10 12 14Quarters

Priv_consumption

-50

510

Per

cent

0 2 4 6 8 10 12 14Quarters

Investment

-10

-50

5P

erce

nt

0 2 4 6 8 10 12 14Quarters

RER

-6-4

-20

24

Per

cen

t

0 2 4 6 8 10 12 14Quarters

Exports-5

05

10P

erce

nt

0 2 4 6 8 10 12 14Quarters

Imports

21

two years after the shock. In sharp contrast, monetary authorities raise interest

rates under IT; albeit only mildly and with some lag in response to the pickup in

prices. For fiscal policy this pattern is reversed. While targeters accommodate

the shock, non-inflation targeters reduce fiscal spending. We interpret these

findings in the light of a coordinating role of IT for monetary and fiscal policy.

While it is undisputed that IT requires the absence of fiscal dominance (Sargent

and Wallace 1981, Freedman and Ötker-Robe 2010), Minea and Tapsoba (2014)

document that the adoption of IT improves the fiscal discipline of developing

countries. Thus, a higher coefficient on inflation in the loss function of central

banks enforces a sound fiscal position which, in turn, prevents from pro-cyclical

public spending in the wake of large disasters.

Next to public, private spending seems to explain a relevant share of the

difference in the output responses between regimes. Finally, regarding the

external sector, there is some evidence that targeters benefit from a more stable

real exchange rate, which tends to appreciate in the other group where

consumer prices increase sharply.

Figure 5: Differential level effects of large real shocks between targeting and non-inflation targeting economies

Note: The figure shows the difference between inflation targeting and non-inflation targeting countries with respect to the response of the level (cumulated first (log) difference) of key macroeconomic variables to large natural disasters over the period 1970Q1-2015Q4. Statistical inference is based on 500 Monte-Carlo draws.

01

23

4P

erce

nt

0 2 4 6 8 10 12 14Quarters

GDP

-15

-10

-50

Per

cent

0 2 4 6 8 10 12 14Quarters

Prices

01

23

Per

cent

0 2 4 6 8 10 12 14Quarters

Policy_rate

-20

24

6P

erce

nt

0 2 4 6 8 10 12 14Quarters

Gov_consumption

02

46

Per

cent

0 2 4 6 8 10 12 14Quarters

Priv_consumption

-10

-50

510

Per

cent

0 2 4 6 8 10 12 14Quarters

Investment

-50

51

0P

erce

nt

0 2 4 6 8 10 12 14Quarters

RER

-4-2

02

46

Per

cen

t

0 2 4 6 8 10 12 14Quarters

Exports

-10

-50

5P

erce

nt

0 2 4 6 8 10 12 14Quarters

Imports

22

The cumulative differential effects between targeters and non-inflation

targeters, which allow for a more precise estimation and quantification of the

dynamic effects of IT, are presented in Figure 5. They add to the evidence in

favor of the first hypothesis as they underline the superior performance of IT

economies. GDP is significantly higher under IT in the quarter of impact and

subsequently. After four years, the difference is roughly 2 percentage points. At

the same time, the increase in consumer prices is more than 5 percentage points

lower. As indicated earlier, some of these marked differences in the behavior of

output and prices across monetary regimes seem to be attributable to the direct

effects of divergent monetary and fiscal policies across regimes, as well as to the

differing paths of private consumption. After four years, the level responses of

the respective variables differ by 2, 2, and 4 percentage points, respectively,

between regimes.

To formally evaluate our first hypothesis we compute the average inflation

and output growth rate for targeting and non-inflation targeting economies over

the response horizon of four years and test whether the means are different

across country groups. As the underlying responses are random vectors with

distributions, we first investigate the precision and distribution of our estimates

of average inflation and output growth, following Cecchetti and Rich (2001).

Figure A3 in the Online Appendix plots the empirical density functions of the

estimates obtained from the Monte Carlo simulations.11 Based on the properties

of the empirical densities, we proceed by estimating the means of these

distributions and testing whether they are significantly different across regimes.

Table 1 contains the results for output growth and inflation, as well as for the

other variables shown in the impulse responses. The first two columns lend

further support to the first hypothesis. The average quarterly rate of output

growth following a shock is 0.16 percentage points higher under IT, and the

average quarterly change in the price level is 0.44 percentage points lower.

These differences are statistically highly significant according to the

corresponding t-statistics and p-values. The results for the other variables are

11

Several observations are worth mentioning. First, the figure corroborates the conclusion based on the impulse response analysis of a significantly better macroeconomic performance of targeting economies, namely higher output growth and lower inflation. For both variables, the distributions overlap only marginally. Second, it shows that the effects for non-targeters are less precisely estimated as the distributions are less concentrated. Third, it indicates that it is reasonable to assume that the true means, which are nonlinear functions of normally distributed variables, are also normally distributed.

23

mostly equally stark and confirm the two reasons for the superior growth and

inflation performance of targeters indicated by the impulse response analysis: a

different policy mix and better shock absorption through the external economy.

All in all, we conclude that IT significantly reduces inflation and increases

output growth when an economy is subject to large real shocks.

Table 1. Testing for differences in means

Variable D.

GDP D.

Prices D.

Pol. rate D.

Gov. cons. D.

Priv. cons. D.

Investm. D.

RER D.

Exports D.

Imports

IT

0.07 0.05 0.03 0.04 0.09 0.19 -0.04 -0.01 0.09

non-IT -0.09 0.48 -0.09 -0.16 -0.17 0.12 -0.22 -0.21 0.15

Difference 0.16 -0.44 0.12 0.2 0.26 0.07 0.18 0.2 -0.07

t-statistic 51.62 -40.9 68.98 44.38 59.31 6.15 12.73 26.76 -7.01

p-value 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00

Notes: The table shows the estimated mean of the (log) change of main macroeconomic variables over four years following natural disasters in inflation targeting and non-inflation targeting economies, as well as the differences between the means together with their t-statistics and p-values based on 500 Monte Carlo draws.

3.3. Macroeconomic volatility and financial frictions

The finding that IT generates both higher output growth and lower inflation is

remarkable given that there is also a contention in the literature whether IT can

only reduce inflation at the expense of depressing output (Cecchetti and Rich,

2001; Friedman, 2004; Gonçalves and Carvalho, 2009). Beyond that, the direct

effects of the different policy mix adopted by targeting economies seem not to

reveal the full story behind the success. There are additional features of IT,

however, that are prominently discussed in the literature and which are thought

to contribute to its superiority over alternative monetary regimes (Bernanke

and Mishkin, 1997): (i) the attainment of a generally more stable economic

environment, and (ii) a reduction of financial frictions. In this section, we show

empirical evidence in the form of conditional effects of inflation targeting in

support of these two aspects of the monetary policy regime.

To test the first argument, we assess the effect of IT on macroeconomic

volatility. For this, we again rely on the distributions of the estimated impulse

responses and compute, analogously to the procedure for mean growth rates,

for each variable the distribution of the standard deviation of its growth rate

24

over the response horizon of four years. With these distributions and the

implied average standard deviations at hand, we can formally evaluate our

second hypothesis by testing whether IT reduces the variability of inflation and

output growth in the presence of large real shocks.

Table 2. Testing for differences in volatility

Variable D.

GDP D.

Prices D.

Pol. rate D.

Gov. cons. D.

Priv. cons. D.

Investm. D.

RER D.

Exports D.

Imports

IT

0.16 0.15 0.07 0.29 0.13 0.46 0.64 0.48 0.44

non-IT 0.28 0.45 0.23 0.66 0.36 1.16 1.14 1.09 1.05

Difference -0.12 -0.30 -0.16 -0.37 -0.23 -0.70 -0.50 -0.61 -0.61

t-statistic -46.99 -64.52 -92.04 -53.66 -79.55 -66.95 -50.16 -69.29 -66.17

p-value 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00

Notes: The table shows the estimated average standard deviation of the (log) change of main macroeconomic variables over four years following a large real shock in inflation targeting and non-inflation targeting economies, as well as the differences between the mean standard deviations together with their t-statistics and p-values based on 500 Monte Carlo draws.

Table 2 presents evidence in favor of argument (i). All standard deviations are

significantly lower under IT. This observation holds for both nominal and real

variables as well as for the domestic and external sector. The differences are all

highly statistically significant.12 Regarding the domestic economy, the standard

deviation of inflation is roughly 30 percentage points lower under IT, and that of

output growth is about 10 percentage points smaller. Based on these results, we

accept the second hypothesis. Regarding the external economy, lower volatility

of real exchange rate growth is coupled with lower fluctuations in export and

import growth. Together with the model-consistent superior output growth

performance under IT documented in the previous section, these differences in

volatility lend empirical support to the idea that IT supports output growth by

establishing a generally more stable macroeconomic environment. This stability

could also influence the degree of financial frictions in an economy.

We focus on two particular types of frictions to evaluate argument (ii): credit

risk premia and term premia. To see whether the behavior of these premia is

affected by the monetary regime, we study the dynamic behavior of different

12 The empirical density functions for the estimated standard deviations are shown in Figure A4 in the Online Appendix. As with the density functions for estimated means, they all appear to be well, that is, normally shaped.

25

private and public interest rates. The first panel of Error! Reference source

notfound. repeats the differential response of the policy rate between targeting

and non-inflation targeting regimes for comparison. The following panels

contain the differences in the behavior of the short-term Treasury bill rate, the

money market rate, and the lending rate of the private sector.13

Figure 6: Differential response of monetary policy and interest rates to large natural disasters between targeting and non-inflation targeting economies

Note: The figure shows the difference between the responses of public and private interest rates in inflation targeting and non-inflation targeting economies to large natural disasters over the period 1970Q1-2015Q4. Statistical inference is based on 500 Monte-Carlo draws.

The panels show that there are substantial differences in the pass-through of

monetary interventions from the public to the private sector across regimes.

While the difference between the response of the policy and the T-Bill rate is

roughly 2 percentage points on average over four years, this gap narrows to 1.5

percentage points for the money market rate and to 1 percentage point for the

lending rate. In the latter case, the difference is also less statistically significant.

These numbers and underlying interest rate dynamics suggest that larger

countercyclical private credit risk premia reduce the effectiveness of monetary

13 The underlying level responses for targeting and non-inflation targeting economies are deferred to Figure A5 in the Online Appendix.

01

23

Per

cent

0 2 4 6 8 10 12 14Quarters

Policy_rate-2

02

46

Per

cent

0 2 4 6 8 10 12 14Quarters

Tbill_rate

-10

12

34

Per

cent

0 2 4 6 8 10 12 14Quarters

Money_rate

-10

12

3P

erce

nt

0 2 4 6 8 10 12 14Quarters

Lending_rate

-2-1

01

23

Per

cent

0 2 4 6 8 10 12 14Quarters

Long_rate

05

10

15

20P

erce

nt

0 2 4 6 8 10 12 14Quarters

Cumul_policy_rate

-2-1

01

23

Per

cen

t

0 2 4 6 8 10 12 14Quarters

Real_rate

05

10

15

20

25P

erce

nt

0 2 4 6 8 10 12 14Quarters

Cumul_real_rate

-15

-10

-50

Per

cen

t

0 2 4 6 8 10 12 14Quarters

Prices

26

policy under non-inflation targeting regimes. This seems to partially explain

their poorer output performance notwithstanding aggressive monetary easing.

On the other hand, the results indicate that the behavior of public credit risk

premia is similar across regimes and thus does not help explain the different

responses of fiscal policy as the response of the T-Bill rate is quantitatively

similar to that of the policy rate.

What is striking nevertheless, is that despite the sharp relative increase of the

T-Bill rate in targeting economies, there is no significant difference in the path

of long-term rates (see middle panel). Together, these two observations suggest

that the term premium responds quite different across monetary regimes. The

expectations hypothesis of the term structure in its linear form implies that the

nominal long-term rate is the sum of the path of the current and future expected

nominal short-term rates and the term premium. To obtain a measure of the

expected short rates, we compute and plot the cumulative difference in the

policy rate, which can also be interpreted as the model-based differential

change in the yield curve between regimes in the quarter of impact. The sharp

relative steepening of the curve can only be consistent with an essentially

unchanged difference in the long rate if the difference between the term premia

in targeting and non-inflation targeting economies strongly declines. Indeed, the

group-specific responses suggest that the term premium remains roughly

constant under IT, whereas it sharply increases otherwise, such that the

differential term premium drops (see Figure A5 in the Online Appendix).

We reach similar conclusions when looking at the behavior of expected real

rates and inflation. The standard term structure theory further implies that the

nominal long rate is the sum of current and future expected real short rates,

expected inflation, and the term premium. To obtain a measure of expected real

short rates we first estimate the response of current real rates, computed as the

difference between the policy rate and realized inflation, and then cumulate the

response. To approximate expected inflation, we employ the model-based

change in the price level. As the figure shows, the relative increase in the

expected real rates for targeters is quantitatively not fully compensated by

relative declines in inflation expectations, which implies that the relative term

premium must decline for the difference in the nominal long rate to remain

27

stable. Altogether, the results of the subsection help to understand why a strong

monetary stimulus coupled with a mild fiscal contraction in non-inflation

targeting economies yields significantly inferior outcomes than a moderate

monetary tightening with fiscal accommodation under IT.

4. Sensitivityanalysis

In this section we perform various sensitivity tests to see whether our main

results are robust. We (i) check whether sample splits change the main results

and (ii) control for other potential shock absorbers In the Online Appendix, we

further consider modified versions of the shock and IT measures, and use an

alternative estimator as well as different model specifications.

Figure 7: Differential responses between targeting and non-inflation targeting economies in subsamples

Note: The figure shows for alternative subsamples the differential response of the level (cumulated first (log) difference) of GDP, prices and the policy rate across targeting and non-inflation targeting countries to large natural disasters over the period 1970Q1-2015Q4. Row (1) shows the baseline results using the full sample for comparison. In row (2), the sample is restricted to OECD member countries. In row (3), the sample is restricted to non-OECD member countries. Statistical inference is based on 500 Monte-Carlo draws.

01

23

4P

erce

nt

0 2 4 6 8 10 12 14Quarters

GDP

-15

-10

-50

Per

cent

0 2 4 6 8 10 12 14Quarters

Prices

01

23

Per

cent

0 2 4 6 8 10 12 14Quarters

Policy_rate

Ful

l sa

mpl

e

01

23

45

Per

cent

0 2 4 6 8 10 12 14Quarters

GDP

-20

-15

-10

-50

Per

cent

0 2 4 6 8 10 12 14Quarters

Prices

01

23

4P

erce

nt

0 2 4 6 8 10 12 14Quarters

Policy_rate

OE

CD

-10

-50

510

Per

cent

0 2 4 6 8 10 12 14Quarters

GDP

-60

-40

-20

020

Per

cent

0 2 4 6 8 10 12 14Quarters

Prices

-10

-50

5P

erce

nt

0 2 4 6 8 10 12 14Quarters

Policy_rate

non-

OE

CD

28

4.1. Subsample estimates

We split the sample to find out whether certain country groups are driving the

main results. In particular, we divide the sample into developed and developing

countries. The reason is that on the one hand, richer economies might be more

likely to adopt IT, given their typically more developed democratic and financial

institutions, and at the same time to weather large natural disasters better. On

the other hand, there is evidence in the literature that the introduction of IT has

in particular an impact on economic performance in developing economies

(Ball, 2010; De Mendonça and e Souza, 2012). Lin and Ye (2007), for example,

find no effect of IT in seven industrial countries, whereas Lin (2010) detects a

significant impact of IT in developing countries.

We re-estimate model (2) for two subsamples. First, we look at countries that

are members of the OECD. Second, we look at the complement subsample of

countries which are not OECD members. Figure 7 shows the differential level

responses under IT and non-IT regimes in the subsample. The baseline results

for the full sample are in the first row for comparability. The dynamic responses

of the OECD subsample are qualitatively and quantitatively similar to the

baseline results. There is higher GDP growth, prices increase less, and monetary

policy is more restrictive. This picture changes somewhat for the subsample of

non-OECD countries. The difference in output performance largely vanishes, but

prices remain lower in IT countries despite the initial relative easing of

monetary policy. We conclude that both developed and developing economies

tend to benefit from an improved macroeconomic performance under IT, but

the baseline results seem to be mainly driven by the OECD sample.

4.2. Does hard or soft targeting make a difference?

As a next step in the robustness analysis we try to determine whether the

adoption of IT per se generates macroeconomic improvements. Such a finding

would speak against the “conservative window-dressing” view which postulates

that the very features of IT have little or no effects on output or inflation and

instead the stronger emphasis of the central bank on inflation and the

corresponding conduct of monetary policy achieves the better outcomes

(Romer, 2006).

29

Figure 8: Duration of missed inflation targets

Note: PANEL(A): Density of average duration spells where the inflation rate is outside the target corridor in the sample of countries operating under an inflation targeting regime. “Target misses” are defined as observed CPI inflation rates outside of the target corridor. The solid line represents the kernel density estimate of a Gaussian kernel function with a bandwidth of 2 and 0.05, respectively. PANEL (B): Density plot of the longest time period of a one-sided, continued realized inflation rates outside of the corridor per country in the sample operating under IT. The maximum duration of a missed inflation target is expressed in percent of the total number of quarters under IT.

To test this argument we split the IT group into a hard targeting group that ex

post complies more strictly with the inflation target versus a soft targeting

group that ex post complies less with the inflation target. We measure

compliance with the inflation target as the maximum time spell of consecutive

recordings of inflation rates outside of the target corridor.14 Figure 8a shows the

histogram of the average duration of target misses in the sample. There is no

country with an average duration of target misses at zero or one quarter and the

highest density is at two quarters, rapidly declining to 14 quarters. There are

outliers of up to 27 quarters of continued misses. Figure 8b exhibits the

maximum one-sided duration spell of each IT country. It is expressed in percent

over the total number of quarters under IT. This is the measure we use to

separate hard from soft inflation targeters. We split the sample according to the

50th percentile of the maximum duration spell of target misses. This leads to a

threshold value of 11.4 percent. Thus, the countries with a maximal one-sided

deviation from target of more than 11.4 percent of the total number of quarters

under IT are declared as soft IT counties.

14 We compiled a database for the inflation targeting countries and their respective target rates and target corridors. Where no corridors are used for the conduct of IT, we constructed a symmetric and hypothetical corridor around the target rate with the

average size of target corridors across countries. Figure A6 and Figure A7 in the Online Appendix illustrate this for the entire sample of IT countries.

010

2030

Per

cen

t

0 5 10 15 20 25 30Duration in quarters

02

46

810

Fre

quen

cy

0 .1 .2 .3 .4 .5 .6 .7 .8 .9 1Max. duration of missed target

A. Average duration of spells of missed target B. Max. duration of missed target

30

Figure 9: Does hard or soft inflation targeting improve performance?

Note: The figure shows the difference between the responses of the level (cumulated first (log) difference) of all inflation targeting, only less-complying IT countries, and only complying IT countries, respectively, versus all non-inflation targeting economies to large natural disasters over the period 1970Q1-2015Q4. Statistical inference is based on 500 Monte-Carlo draws.

Our classification follows the idea that temporary deviations from the inflation

target are fully in line with an inflation target which should be reached over the

medium term, while different shocks drive the actual inflation rate occasionally

out of the target range. In fact, there is no country in our sample for which

inflation has never deviated from the target. Depending on the size of the shock,

this might also occur by a substantive amount. Further, monetary policy moves

inflation only with some lag, which gives rise to some persistence in the

deviation from target. However, in order to maintain credibility in the overall

target, a prolonged one-sided deviation should result in an enhanced effort by

the central bank to restore the target (Roger and Stone, 2005). We thus think

that the maximum duration that the inflation rate was out of the target range is

a good proxy for the commitment of the central bank to maintaining and

defending the inflation target.

The first row of Figure 9 repeats the baseline results for output, prices, and the

policy rate for comparison. They show the differential responses of all targeters

versus all non-targeters. The middle row contains the differential effects of only

02

46

Pe

rcen

t

0 2 4 6 8 10 12 14Quarters

GDP

-15

-10

-50

Pe

rcen

t

0 2 4 6 8 10 12 14Quarters

Prices

01

23

4P

erc

ent

0 2 4 6 8 10 12 14Quarters

Policy_rateA

ll IT

-20

24

68

Per

cent

0 2 4 6 8 10 12 14Quarters

GDP

-30

-20

-10

010

Per

cent

0 2 4 6 8 10 12 14Quarters

Prices

-20

24

Per

cent

0 2 4 6 8 10 12 14Quarters

Policy_rate

soft

IT

02

46

Per

cent

0 2 4 6 8 10 12 14Quarters

GDP

-15

-10

-50

Per

cent

0 2 4 6 8 10 12 14Quarters

Prices

01

23

4P

erce

nt

0 2 4 6 8 10 12 14Quarters

Policy_rate

hard

IT

31

soft IT versus all non-IT economies and the last row those of only hard IT

against all non-IT regimes. The difference to non-targeters are most pronounced

for hard targeters. For the response of GDP, there are almost no significant

differences between how soft targeting and non-inflation targeting economies

whether large adverse shocks. For the price level and policy response, the

baseline results are maintained in both cases. However, for soft IT the difference

in the price level and policy response is statistically only borderline significant

and the difference in the latter response is quantitatively also smaller. Together,

the findings suggest that the baseline results are mostly driven by hard

targeters and that it is the actual conduct of monetary policy that matters for

successful macroeconomic stabilization.

4.3. Controlling for other potential shock absorbers

We estimate modified versions of model (2) in order to control for alternative

channels that potentially affect the response to a natural disaster shock.

Specifically, we make two extensions. First, we add a second level and

interaction term that accounts for alternative shock absorbing country

characteristic. Second, we add triple-interaction terms which allow testing

whether there is a marginal effect of running an IT regime conditional on other

country features. In the first approach, we extend the model with a generic

additional variable 4�,��� as follows:

We reduce the impulse horizon to three years to reduce the number of

additional parameters to be estimated. Moreover, in most specifications we

introduce the alternative shock absorbers one-by-one, and only in final

specification include several of them simultaneously.

We first reconsider the question of whether the level of development matters

for the results by including an interaction between the shock and GDP per capita

in 1997Q1.15 This measure is used as a proxy for the average degree of

15 In this case, we leave out the level term as it is constant over time and hence captured by the country fixed effects.

��,� = � + �'����,��� + (��&�,�����,��� + )��&�,��� + 5�4�,�����,��� + 6�,���4�,���*�

���

+� ���,���

�

���+ ���,��� + � + � + ��,� .

(4)

32

development of an economy, which may affect its capabilities to adjust to

natural disaster shocks. The results are shown in Table 3. We report the

difference in the estimated average first and second moment of GDP growth and

inflation between targeting and non-inflation targeting regimes over the

response horizon, and the corresponding t-statistics obtained from Monte Carlo

simulations. Output growth is significantly higher, inflation is lower, and both

output growth and inflation variability are lower under IT. The differences

remain highly statistically significant and in most cases are larger in absolute

value than in the baseline specification, which is repeated in the first row for

comparison.

Table 3: Controlling for other shock absorbers

Difference mean

difference standard deviation

D.GDP D.Prices

D.GDP D.Prices

Specification Differential effect of IT

Boot-strapped

t-stat

Differential effect of IT

Boot-strapped

t-stat

Differential effect of IT

Boot-strapped

t-stat

Differential effect of IT

Boot-strapped

t-stat

(1) Baseline 0.16 51.62 -0.44 -40.9

-0.12 -46.99 -0.30 -64.52

(2) GDP p.c. 0.15 42.07 -0.72 -55.80

-0.19 -56.99 -0.44 -53.35

(3) Institutions 0.20 46.85 -0.70 -43.82

0.18 39.24 0.30 31.33

(4) FX regime 0.10 10.29 -0.18 -6.15

0.47 50.83 0.32 18.84

(5) Frequency 0.09 22.44 -0.55 -38.92

-0.24 -57.85 -0.45 -55.40

(6) Island 0.14 38.17 -0.65 -49.27

-0.15 -46.43 -0.31 -44.44

(7) Size (km2) 0.15 42.44 -0.50 -45.17

-0.09 -28.24 -0.25 -44.78

(8) GDP size 0.12 32.96 -0.70 -57.28

-0.16 -48.46 -0.35 -49.26

(9) Gov. size 0.24 54.54 -0.54 -38.40

-0.24 -57.83 -0.44 -49.61

(10) Combined 0.08 7.63 -0.19 -6.00

0.14 10.51 -0.19 -6.77

Notes: The table shows the difference between the average estimated mean and standard deviation of GDP growth and inflation over three years following a large real shock in inflation targeting and non-inflation targeting economies, together with their t-statistics based on 500 Monte Carlo draws and when controlling for other potential shock absorbers.

As a further cross-check of whether the level of development matters for our

results, we use the polity IV variable as 4���� since advanced economies usually

have better institutions. This variable is also motivated by the results of Noy

(2009), who finds that the quality of institutions reduces the impact of natural

disasters on aggregate growth. In row (4), we instead control for the exchange

rate regime, as Ramcharan (2007) shows that flexible exchange rates are

conducive to the adjustment to real shocks. Specifically, we use a dummy

variable which is equal to one in case of a fixed exchange rate regime, and zero

33

in case of flexible exchange rates.16 In rows (5)-(7) we replace the exchange rate

dummy with the unconditional frequency by which a country is hit by shocks, an

island dummy, and the size of a country in squared km, respectively, following

Ramcharan (2007). All three interaction terms capture geographic

characteristics of a country that potentially affect both the choice of the

monetary regime and the response to the shock. In rows (8) and (9) we instead

use GDP and government size, respectively, as measures of the ability of a

country to provide internal assistance to disaster-hit regions. In the vast

majority of specifications the main results hold and the effects of IT on first and

second moments are similar to the differences in the baseline specification.

In the last row we include several of the control interaction terms

simultaneously. Specifically, we combine one measure of development (GDP

p.c.) with the exchange rate regime dummy, the frequency, the island dummy,

and one size measure (size in km2). The selection combines the types of shock

absorbers previously investigated. Given the large number of additional

parameters, we reduce the impulse horizon to two years. The estimated effect of

IT on first moments is smaller but significant. The impact on second moments is

mixed. We conclude that while other factors also play a role in the absorption of

shocks, they are not the main explanation for the differential responses between

targeting and non-inflation targeting economies shown in the previous section.

As a second set of robustness checks, we run a model with a triple-interaction

term between the shock, the inflation targeting dummy, and a third variable of

interest of the form

The focus is on the coefficients of the triple-interaction 7�,���. These estimates

capture the marginal effect of inflation targeting that arises due to positive or

negative synergies of this monetary regime with other country characteristic.

16 We use the measure of Reinhart and Rogoff (2004) and map their classification which describes exchange rate regimes on the interval (1,6) into the exchange rate regime dummy variable according to Ramcharand (2007). Specifically, regimes ≤3 are classified as fixed (dummy =1), while 4 and 6 are classified as flexible (dummy=2). We exclude 5=freely falling from the sample.

��,� = � + �[����,��� + (��&�,�����,��� + )��&�,��� + 5�4�,�����,��� + 6�,���4�,���

�

���

+:��&�,���4�,��� + 7�,����&�,���4�,�����,���]

+� ���,���

�

���+ ���,��� + � + � + ��,� .

(5)

34

They are shown in cumulative terms in Figure 10. As the number of coefficients

to be estimated in model (5) increases by two times J, we set J=11.

Figure 10: Controlling for alternative potential shock absorbers

Note: The figure shows the cumulative coefficients of the triple interaction term in model (5) between inflation targeting and other potential shock absorbing country characteristics with large natural disaster shocks. Statistical inference is based on 500 Monte-Carlo draws.

Row (1) depicts a case where 4�,��� is the exchange rate regime.17 There has

been recently renewed interest in the question whether inflation targeting

countries should also let their currencies float freely, or whether it is beneficial

for them to intervene through sterilized trades in order to reduce exchange rate

volatility (Ghosh et al., 2016). The rationale behind the use of targeted

interventions in the foreign exchange market for IT countries is that this

additional instrument can support financial stability in the presence of

unhedged exposure to foreign currency of domestic firms. Ebeke and Fouejieu

(2015), for example, document that there is cross-country heterogeneity with

respect to the exchange rate regime depending on the exposure to foreign

exchange risk.

17 We used the indicator variable of Shambaugh (2004) who codes countries that de facto peg their currencies as a one.

-20

-10

010

20P

erce

nt

0 2 4 6 8 10Quarters

GDP

-50

050

100

Per

cen

t

0 2 4 6 8 10Quarters

Prices

-15

-10

-50

510

Per

cen

t

0 2 4 6 8 10Quarters

Policy_rate

FX

reg

ime

-15

-10

-50

5P

erce

nt

0 2 4 6 8 10Quarters

GDP-4

0-2

00

2040

Per

cent

0 2 4 6 8 10Quarters

Prices

-10

-50

5P

erce

nt

0 2 4 6 8 10Quarters

Policy_rate

G7

-10

-50

51

0P

erce

nt

0 2 4 6 8 10Quarters

GDP

-40

-20

02

04

0P

erce

nt

0 2 4 6 8 10Quarters

Prices

-50

51

0P

erce

nt

0 2 4 6 8 10Quarters

Policy_rate

OE

CD

35

Rows (2) and (3) feature cases when the triple-interaction is constructed with

a G7 or OECD dummy, respectively. These are variations of the previous

hypothesis that the level of development is important for the macroeconomic

benefits of IT. If the effect of IT on, say, GDP growth is larger in non-G7 countries

than in this group, the cumulative triple-interaction terms would be

significantly negative. In all three rows, the cumulative triple-interaction effects

are largely insignificant. One exception is the policy rate in the case of a triple-

interaction with the OECD dummy. The positive effect indicates that OECD

countries that operate under IT raise rates more after large real shocks than

non-OECD countries under IT.

5. Conclusions

In this paper we present robust empirical evidence for the hypothesis that

inflation targeting leads to better economic outcomes. When hit by large

adverse shocks in the form of natural disasters, economies with an inflation

targeting regime experience significantly lower inflation and inflation variability

than under alternative monetary policy regimes. At the same time they enjoy

higher and more stable output growth. We show that the success of inflation

targeting rests on a number of pillars.

First, predominantly hard targeting stabilizes the economy, while soft

targeting has only limited effects on macroeconomic dynamics. Second, our

results suggest that a tougher stance on inflation does not only reduce

consumer price fluctuations but also real exchange rate movements, which

translates into a better adjustment of the economy through the external sector.

Third, the findings indicate that IT, by reducing also the volatility of public and

private interest rates, increases the effectiveness of monetary policy by

lowering credit risk and term premia. Moreover, private consumption and

investment are more stable under IT, which is supportive of growth. Finally, the

adoption of IT with its focus on price stability appears to be coupled with a

stronger orientation of fiscal policy towards output stabilization. All in all, our

findings show that inflation targeting is well and alive and rationalize the

remarkable success of this monetary regime which, once adopted, has never

been abandoned (Rose, 2007; 2014).

36

The paper contributes to the literature by analyzing the different economic

outcomes under alternative monetary policy regimes conditional on large

natural disasters, which are exogenous to the choice of the monetary regime.

This approach to the question is novel as the existing literature has focused on

the unconditional effects of inflation targeting (Walsh, 2009; Ball, 2010). The

departure from looking at the average effect also explains why OECD member

countries in the sample are benefitting more from IT than non-OECD countries,

since the former are more often in the hard-IT group.

The findings of the paper have a number of implications for central banks.

They show that while IT may not strictly be a superior policy mandate in open

economies in normal or tranquil times - as shown by the existing literature - it is

a better mandate in crisis times, at least when the domestic economy is hit by a

large real shock, such as a natural catastrophe. The better adjustment to large

natural disasters of IT countries suggests that IT has been more of a savior

during the Global Financial Crisis than thought. Even if the endogeneity problem

makes it hard to study the role of IT during the GFC, the results obtained

through the analysis of exogenous natural disasters might well be transferable

to the case of the GFC.

However, this holds only if a central bank does not merely pretend to follow

an IT strategy, but if it has gained credibility through a successful track record

on IT. Therefore, it should be considered in the debate on reforming the present

IT frameworks toward more flexibility that allowing for prolonged deviations

from the target range to pursue leaning against the wind policies comes at a cost

in terms of lower shock resilience.

37

REFERENCES

Andersen, T. B., N. Malchow-Møller and J. Nordvig. 2015. “Inflation targeting and macroeconomic performance since the Great Recession”, Oxford Economic

Papers 67(3), 598-613.

Ball, L. 2010. “The Performance of Alternative Monetary Regimes”, Handbook of

Monetary Economics, Vol. 3, 1303-1343.

Ball, L. M. and N. Sheridan. 2004. “Does inflation targeting matter?”, in: Bernanke, B. and M. Woodford (Eds.), The Inflation-Targeting Debate: University of Chicago Press, 249-282.

Beck, N. and J. N. Katz. 1995. “What to do (and not to do) with time-series cross-section data”, American Political Science Review 89(03), 634-647.

Benson, C. and E. J. Clay. 2004. “Understanding the economic and financial impacts of natural disasters”, Disaster Risk Management Series, No. 4. Washington, D.C.: World Bank.

Bernanke, B. S. and F. S. Mishkin. 1997. “Inflation Targeting: A New Framework for Monetary Policy?”, Journal of Economic Perspectives 11(2), 97-116.

Borio, C. 2014. “Monetary policy and financial stability: what role in prevention and recovery?”, BIS Working Papers, No. 440.

Brito, R. D. 2010. “Inflation targeting does not matter: Another look at OECD sacrifice ratios”, Journal of Money, Credit and Banking 42(8), 1679-1688.

Carvalho Filho, I. E. 2010. “Inflation targeting and the crisis: An empirical assessment”, IMF Working Papers, WP/10/45.

Cavallo, E., S. Galiani, I. Noy and J. Pantano. 2013. “Catastrophic natural disasters and economic growth”, Review of Economics and Statistics 95(5), 1549-1561.

Cecchetti, S. G. and R. W. Rich. 2001. “Structural estimates of the US sacrifice ratio”, Journal of Business & Economic Statistics 19(4), 416-427.

Clarida, R., J. Gali and M. Gertler. 1999. "The science of monetary policy: A new Keynesian perspective", Journal of Economic Literature 37(Dec1999), 1661-1707.

Coibion, O. (2012). “Are the effects of monetary policy shocks big or small?”, American Economic Journal: Macroeconomics 4(2), 1-32.

De Mendonça, H. F. and G. J. d. G. e Souza. 2012. “Is inflation targeting a good remedy to control inflation?”, Journal of Development economics 98(2), 178-191.

Ebeke, C. and A. Fouejieu. 2015. “Inflation Targeting and Exchange Rate Regimes in Emerging Markets”, IMF Working Papers, WP/15/228.

Felbermayr, G. and J. Gröschl. 2014. “Naturally negative: The growth effects of natural disasters”, Journal of Development Economics 111, 92-106.

Frankel, J. 2012. “The death of inflation targeting”, VoxEU. org, 19 June 2012.

Frankel, J. A., C. A. Vegh and G. Vuletin. 2013. "On graduation from fiscal procyclicality", Journal of Development Economics 100, 32-47.

38

Freedman, C. and I. Ötker-Robe. 2009. "Country experiences with the introduction and implementation of inflation targeting", IMF Working Paper, WP/09/161.

Friedman, B. M. 2004. “Why the Federal Reserve should not adopt inflation targeting”, International Finance 7(1), 129-136.

Ghosh, A. R., J. D. Ostry and M. Chamon. 2016. “Two targets, two instruments: Monetary and exchange rate policies in emerging market economies”, Journal

of International Money and Finance 60, 172-196.

Gonçalves, C. E. S. and A. Carvalho. 2009. “Inflation targeting matters: Evidence from OECD economies” sacrifice ratios”, Journal of Money, Credit and Banking 41(1), 233-243.

Hallegatte, S. and P. Dumas. 2009. “Can natural disasters have positive consequences? Investigating the role of embodied technical change”, Ecological Economics 68(3), 777-786.

Inoue, H. and Y. Todo. 2017. “Propagation of negative shocks through firm networks: Evidence from simulation on comprehensive supply-chain data”, RIETI Discussion Paper Series, 17-E-044.

Judson, R. A. and A. L. Owen. 1999. “Estimating dynamic panel data models: a guide for macroeconomists”, Economics Letters 65(1), 9-15.

Keen, B. D. and M. R. Pakko. 2011. “Monetary Policy and Natural Disasters in a DSGE model”, Southern Economic Journal 77(4), 973-990.

Kilian, L. 2008. “Exogenous oil supply shocks: How big are they and how much do they matter for the US economy?”, The Review of Economics and Statistics 90(2), 216-240.

Kim, J. 2011. “Inflation Targeting as Constrained Discretion”, Journal of Money,

Credit and Banking 43(7), 1505-1522.

Kydland, F. E. and E. C. Prescott. 1977. “Rules rather than discretion: The inconsistency of optimal plans”, Journal of Political Economy 85(3), 473-491.

Lin, S. 2010. “On the international effects of inflation targeting”, The Review of

Economics and Statistics 92(1), 195-199.

Lin, S. and H. Ye. 2007. “Does inflation targeting really make a difference? Evaluating the treatment effect of inflation targeting in seven industrial countries”, Journal of Monetary Economics 54(8), 2521-2533.

Lin, S. and H. Ye. 2009. “Does inflation targeting make a difference in developing countries?”, Journal of Development Economics 89(1), 118-123.

Liu, Z., D. F. Waggoner and T. Zha. 2011. “Sources of macroeconomic fluctuations: A regime-switching DSGE approach”, Quantitative Economics 2(2), 251-301.

Loayza, N. V., E. Olaberria, J. Rigolini and L. Christiaensen. 2012. “Natural disasters and growth: Going beyond the averages”, World Development 40(7), 1317-1336.

Minea, A. and R. Tapsoba. 2014. "Does inflation targeting improve fiscal

39

discipline?" Journal of International Money and Finance 40, 185-203.

Nickell, S. 1981. “Biases in dynamic models with fixed effects”, Econometrica 49(6), 1417-1426.

Noy, I. 2009. “The macroeconomic consequences of disasters”, Journal of

Development Economics 88(2), 221-231.

Ramcharan, R. 2007. “Does the exchange rate regime matter for real shocks? Evidence from windstorms and earthquakes”, Journal of International

Economics 73(1), 31-47.

Reichlin, L. and R. Baldwin. (Eds.). 2013. "Is inflation targeting dead? Central banking after the crisis", London: Centre for Economic Policy Research.

Roger, S. 2009. “Inflation Targeting at 20: Achievements and Challenges”, IMF

Working Papers, WP/09/236.

Roger, S. and M. Stone 2005 “On Target? the International Experience with Achieving Inflation Targets”, IMF Working Papers, WP/05/163.

Romer, C. D. and D. H. Romer 2004. “A new measure of monetary shocks: Derivation and implications”, The American Economic Review 94(4), 1055-1084.

Romer, D. 2006. Advanced macroeconomics. New York:McGraw-Hill Companies.

Rose, A. K. 2007. “A stable international monetary system emerges: Inflation targeting is Bretton Woods, reversed”, Journal of International Money and

Finance 26(5), 663-681.

Rose, A. K. 2014. “Surprising similarities: Recent monetary regimes of small economies”, Journal of International Money and Finance 49, 5-27.

Skidmore, M. and H. Toya. 2002. “Do natural disasters promote long-run growth?”, Economic Inquiry 40(4), 664-687.

Smets, F. and R. Wouters. 2007. “Shocks and frictions in US business cycles: A Bayesian DSGE approach”, The American Economic Review 97(3), 586-606.

Sargen, T. J. and N. Wallace. 1981. "Some unpleasant monetarist arithmetic" Federal Reserve Bank of Minneapolis Quarterly Review (Fall 1981), 1-17.

Solow, R. M. 1956. “A contribution to the theory of economic growth”, The

Quarterly Journal of Economics 70(1), 65-94.

Stiglitz, J. E. 2008. “The failure of inflation targeting”, Project Syndicate, 6 May 2008.

Strulik, H. and T. Trimborn. 2014 “Natural disasters and macroeconomic performance: The role of residential investment”, Cege Discussion Papers, No. 194.

Svensson, L. E. 2009. "Flexible inflation targeting", Speech at the Bank for International Settlement at the workshop 'Towards a new framework for monetary policy? Lessons from the crisis', Amsterdam, 21 September.

Svensson, L. E. 2010. “Inflation Targeting”, Handbook of Monetary Economics, Vol. 3, 1237-1302.

40

Taylor, J. B. 2007. “Housing and Monetary Policy”, NBER Working Paper, No. 13682.

Vega, M. and D. Winkelried. 2005. “Inflation targeting and inflation behavior: A successful story?”, International Journal of Central Banking 1(3), 153-175.

Walsh, C. E. 2009. “Inflation targeting: What have we learned?”, International

Finance 12(2), 195-233.

- 1 -

ONLINE APPENDIX.— NOT FOR PUBLICATION

This Online Appendix contains supplementary material for the paper “Inflation

Targeting as a Shock Absorber”, by Marcel Fratzscher, Christoph Große Steffen, and

Malte Rieth.

I. Further sensitivity analysis

I.a. Alternative lag length selection and estimator

We evaluate whether the choice of the lag length of the endogenous variables or the

choice of the estimator changes our main conclusions. The first sensitivity test is

motivated by Coibion (2012), who shows that the size of the effects of monetary policy

estimated by Romer and Romer (2004) depends, among others, on the number of lags of

the endogenous variables. Figure I.1 shows the differential response of output, prices,

and the policy rate between targeting and non-inflation targeting regimes. The first row

contains the results when including three lags of the endogenous variables, instead of

four lags as in the baseline specification. The difference in the effects of IT is

qualitatively not and quantitatively as well as in terms of statistical significance only

mildly affected by this change of the model. We obtain similar results when reducing the

number of lags to two in the next row.

The second sensitivity tests use alternative estimators to address remaining concerns

about biased estimates when using fixed effects and lagged endogenous variables as

regressors (Nickell, 1981), although this bias is known to be small in samples where the

time-dimension is long (T>30, Judson and Owen, 1999), as in our case. In the third row,

we explicitly model the error term structure using panel corrected standard errors,

following Beck and Katz (1995). The error structure accounts for panel

heteroskedasticity and contemporaneously as well as serially correlated errors. Panel

heteroskedasticity and contemporaneous correlation can arise in large cross- country

panels where the level of the endogenous variables differs across countries and where

countries are potentially affected by correlated shocks. In row four, we instead use

feasible generalized least squares estimation with heteroskedastic and first-order

autocorrelated errors. In both cases, we include a full set of country dummies. In the

- 2 -

final row, we return to the within-transformation, but model the autocorrelation in the

error term. In all three cases we report the point estimates together with their 90

percent asymptotic standard errors. All in all, using these alternative estimators does

not change our main results. The differential responses are qualitatively and

quantitatively similar to the baseline results.

Figure I.1: Alternative lag length and estimator

Note: The figure shows the difference between the responses of inflation targeting and non-inflation targeting countries to large

natural disasters based on alternative number of lags of the endogenous variable (row 1 - two lags, row 2 - three lags; 68 and 90

percent confidence bands based on 500 Monte-Carlo draws) and based on alternative estimators (row 3 - panel corrected standard

errors, row - 4 fixed effects with AR(1) error terms, row 5 - feasible generalized least squares with heteroskedastic cross-sectional

and first-order auto-correlated errors; all with 90 percent asymptotic confidence bands).

I.b. Alternative shock selection and modified classification of inflation targeters

Finally, we assess whether the main results are sensitive to the definition of shocks or

inflation targeting. Figure I.2 shows the differential effects across targeting and non-

inflation targeting economies of large disasters on output, prices, and the policy rate.

01

23

4P

erce

nt

0 2 4 6 8 10 12 14

GDP

-10

-50

5P

erce

nt

0 2 4 6 8 10 12 14

Prices

01

23

Per

cent

0 2 4 6 8 10 12 14

Policy_rate

3 la

gs e

ndog

.

01

23

4P

erce

nt

0 2 4 6 8 10 12 14

GDP

-10

-50

5P

erce

nt

0 2 4 6 8 10 12 14

Prices

01

23

Per

cent

0 2 4 6 8 10 12 14

Policy_rate

2 la

gs e

ndog

.

01

23

4P

erce

nt

0 2 4 6 8 10 12 14Quarters

-8-6

-4-2

02

Per

cent

0 2 4 6 8 10 12 14Quarters

01

23

Per

cent

0 2 4 6 8 10 12 14Quarters

Pan

el c

orr.

S.E

.

01

23

4P

erce

nt

0 2 4 6 8 10 12 14Quarters

-10-

8-6

-4-2

0P

erce

nt

0 2 4 6 8 10 12 14Quarters

0.5

11.

52

Per

cent

0 2 4 6 8 10 12 14Quarters

FG

LS

het

ero

sk. A

R(1

)

01

23

4P

erce

nt

0 2 4 6 8 10 12 14Quarters

-10

-50

5P

erce

nt

0 2 4 6 8 10 12 14Quarters

0.5

11.

52

2.5

Per

cent

0 2 4 6 8 10 12 14Quarters

FE

AR

(1)

erro

rs

- 3 -

The first row is based on a subset of the shocks considered in the main analysis.

Specifically, we use only the upper 75th percentile of the damage variable, instead of the

upper 50th percentile in the baseline specification, and from those select the bottom 50th

percentile of GDP deviations, as before. This choice leaves us with 57 shocks under IT

and 23 shocks otherwise. By focusing on even larger disasters, we aim at further

eliminating noise in damage reporting. The first row shows that the differential effects

tend to be similar to the baseline results. Next, we keep the upper 75th percentile of the

damage variable, but leave out the second step in the selection procedure to see whether

conditioning on GDP drops changes the results. In this case, we obtain 112 shocks for

non-inflation targeters and 49 shocks for targeters. The second row shows that the

differential effects of IT are qualitatively unaffected.

Figure I.2: Alternative shock definition and classification of inflation targeters

Note: The figure shows the difference between the responses of inflation targeting and non-inflation targeting countries to large natural disasters based on alternative selections of the shocks (first two rows) and a modified definition of IT (last row). Statistical inference is based on 500 Monte-Carlo draws.

GDP Prices Policy_rate

3 la

gs e

ndog

.

01

23

4P

erce

nt

0 2 4 6 8 10 12 14Quarters

-15

-10

-50

Per

cent

0 2 4 6 8 10 12 14Quarters

0.5

11.

52

2.5

Per

cent

0 2 4 6 8 10 12 14Quarters

75x5

0 pe

rce

nt.

-10

12

3P

erce

nt

0 2 4 6 8 10 12 14Quarters

-15

-10

-50

Per

cent

0 2 4 6 8 10 12 14Quarters

-.5

0.5

11.

5P

erce

nt

0 2 4 6 8 10 12 14Quarters

75th

per

cen

tile

01

23

Per

cent

0 2 4 6 8 10 12 14Quarters

-8-6

-4-2

02

Per

cent

0 2 4 6 8 10 12 14Quarters

0.5

11.

52

2.5

Per

cent

0 2 4 6 8 10 12 14Quarters

Unw

eig

hted

01

23

4P

erce

nt

0 2 4 6 8 10 12 14Quarters

-10

-50

Per

cent

0 2 4 6 8 10 12 14Quarters

0.5

11.

52

Per

cent

0 2 4 6 8 10 12 14Quarters

Sp

illov

ers

01

23

4P

erce

nt

0 2 4 6 8 10 12 14Quarters

-15

-10

-50

Per

cent

0 2 4 6 8 10 12 14Quarters

01

23

Per

cent

0 2 4 6 8 10 12 14Quarters

IT w

/o E

CB

- 4 -

Next, we test whether the weighting of disasters by the onset month affects the main

results. In the third row we use unweighted damages to construct the shocks, whereas in

the fourth row we weigh the reported damage by the onset month, as before, but

additionally take into account up to two months of spillovers in the next quarter. In the

first case, the estimation precision tends to decrease, while it tends to increase in the

second. In both cases, the main results are retained. Last, we use an alternative

classification of IT countries as there is no consensus in the literature whether the ECB is

an inflation targeter nor not (Ball, 2010; Rose, 2014). We thus re-define all 18 euro area

countries as non-inflation targeters, different to the baseline classification where all

countries are coded as targeters when entering the euro. The bottom row shows that the

results are mostly unchanged.

- 5 -

II. SupplementaryTablesandFigures

Table A1-: Inflation targeting adoption dates Country IT adoption date2 euro adoption date

Albania January 2009

Australia1 April 1993

Brazil June 1999

Canada1 February 1991

Chile1 September 1999

Colombia September 1999

Czech Rep1 December 1997

Ghana May 2007

Guatemala January 2005

Hungary1 June 2001

Iceland1 March 2001

Indonesia July 2005

Israel1 June 1997

Korea Rep1 January 2001

Mexico1 January 2001

New Zealand1 January 1990

Norway1 March 2001

Peru January 2002

Philippines January 2002

Poland1 December 1998

Romania August 2005

Serbia September 2006

South Africa February 2000

Sweden1 January 1993

Thailand May 2000

Turkey1 January 2006

United Kingdom1 December 1992

Finland1 February 1993 January 1999

Slovakia1 January 2005 January 2009

Spain1 January 1995 January 1999

Austria1

January 1999

Belgium1

January 1999

Cyprus

August 2008

Estonia1

January 2011

France1

January 1999

Germany1

January 1999

Greece1

January 2001

Ireland1

January 1999

Italy1

January 1999

Latvia1

January 2014

Lithuania

January 2015

Luxembourg1

January 1999

Malta

January 2008

Netherlands1

January 1999

Portugal1

January 1999

Slovenia1

January 2007

Notes: 1) OECD member country. 2) Date of adopting fully fledged inflation targeting. Countries that introduced inflation targeting before entering the euro are coded as targeters from the initial adoption of inflation targeting onwards. Sources: Roger (2009), National central banks, European Central Bank.

- 6 -

Table A2: Country list

Albania India Pakistan

Algeria Indonesia Peru

Argentina Ireland Philippines

Australia Israel Poland

Austria Italy Portugal

Belgium Japan Romania

Brazil Jordan Russia

Bulgaria Kazakhstan Serbia

Canada Kenya Singapore

Chile Korea Rep Slovakia

Colombia Kyrgyzstan South Africa

Croatia Latvia Spain

Cyprus Lithuania Sri Lanka

Czech Rep Luxembourg Sweden

Denmark Malaysia Switzerland

Egypt Malta Thailand

Estonia Mauritius Trinidad and Tobago

Finland Mexico Tunisia

France Mongolia Turkey

Georgia Morocco Ukraine

Germany Namibia United Kingdom

Greece Netherlands United States

Guatemala New Zealand Viet Nam

Honduras Nigeria Yemen

Hungary Norway

Iceland Oman

Notes: The table lists the countries included in the analysis.

- 7 -

Table A3: Data sources Variable Definition Source

DAMw50X50 Damage from disasters in % of GDP: upper 50th percentile of damage reported, lower 50th percentile of deviations from GDP per capita trend growth rate

EM-DAT, IMF-IFS

DAMw50 Damage from disasters in % of GDP: upper 50th percentile of damage reported

EM-DAT, IMF-IFS, OECD, National Sources

itecb Dummy for inflation targeting, including the euro area See Table A1

it Dummy for inflation targeting See Table A1

GDP Real per capita GDP OECD, national sources, WDI

Prices CPI price index, IMF-IFS

Policy rate Core rate at which banks can borrow from the national central bank, end of period, percent

Datastream

Government consumption

Real government consumption, seasonally adjusted OECD, national sources

Private consumption Real private consumption, seasonally adjusted OECD, national sources

Investment Gross capital formation, seasonally adjusted OECD, national sources

REER Real effective exchange rate index BIS

Exports Real exports, seasonally adjusted OECD, national sources

Imports Real imports, seasonally adjusted OECD, national sources

Tbill rate Yield on government bond, 3 month maturity, percent Datastream

Money rate Money market rate, percent IMF-IFS

Lending rate Lending rate, percent IMF-IFS

Long rate Yield on long-term government bond, percent IMF-IFS

Democracy Polity IV, scaled and standardized on the interval [0,1], with 1 indicating a high level of democratic institutions

Center for Systemic Peace

Urbanization Urban population in percent of total population, annual frequency World Bank/WDI

Density Population (thousand) per land area (square kilometers), annual frequency

World Bank/WDI

Notes: The table lists the variables, definitions, and data sources used in the empirical analysis.

- 8 -

Table A4: Probit estimates

(1) (2) (3) (4) (5)

Dependent variable: inflation targeting

Shock(t) 0.002 -0.078 -0.079 -0.077 -0.096 Shock(t-1) 0.034 -0.008 -0.055 -0.048 -0.063 Shock(t-2) 0.030 -0.049 -0.077 -0.058 -0.077 Shock(t-3) 0.028 -0.040 -0.070 -0.047 -0.070 Shock(t-4) 0.023 -0.027 -0.036 -0.017 -0.037 Shock(t-5) 0.021 -0.039 -0.029 -0.010 -0.029 Shock(t-6) 0.035 0.018 0.019 0.027 0.006 Shock(t-7) 0.035 -0.025 -0.031 -0.023 -0.035 Shock(t-8) 0.036 -0.025 -0.028 -0.027 -0.042 Shock(t-9) 0.036 -0.022 -0.028 -0.026 -0.046

Shock(t-10) 0.034 -0.051 -0.053 -0.050 -0.072 Shock(t-11) 0.035 -0.019 -0.020 -0.016 -0.037 Shock(t-12) 0.041 0.014 0.008 0.013 -0.010 Shock(t-13) 0.035 -0.014 -0.027 -0.025 -0.046 Shock(t-14) 0.027 -0.022 -0.041 -0.041 -0.058 Shock(t-15) 0.012 -0.050 -0.056 -0.060 -0.077

Time-invariant characteristics Yes Yes Yes Yes Yes Time-varying characteristics Yes Yes Yes Yes Yes

Lags 1 to 4 inflation Yes Yes Yes Yes Lags 1 to 4 GDP growth Yes Yes Yes

Inflation volatility Yes Yes GDP growth volatility Yes

Observations 10550 4908 4295 4287 4254 Pseudo R-squared 0.28 0.30 0.31 0.32 0.33

p-value F-test shocks 0.999 1.000 0.995 0.999 0.975

Notes: The table shows estimated probit models where the dependent variable is the probability of a country to target inflation. The explanatory variables are natural disaster shocks (lag 0 to 15) and other control variables. Time-invariant characteristics include the number of shocks per country in the sample, the cumulative damage of these shocks per country, per capita GDP in 1997, a dummy variable indicating African countries, and a dummy for islands. Time-varying characteristics contain lag 4 of democracy, the share of urban population, density, country size, and population, respectively. Inflation and GDP growth volatility are measured as the rolling realized variances of these variables over both four and eight quarters. Statistical significance is based on robust standard errors and would be indicated by stars. The shocks are not significant, however, at any lag length. The bottom row of the table shows the p-value of the F-test for joint significance of all lags.

- 9 -

Figure A1: Effects of large real shocks on first (log) differences in targeting and non-inflation targeting economies

Note: The figure shows the response of the first (log) difference of key macroeconomic variables to large natural disasters in inflation targeting (orange line, yellow area) and non-inflation targeting economies (red line, orange area) over the period 1970Q1-2015Q4. Statistical inference is based on 500 Monte-Carlo draws.

-1-.

50

.5P

erc

ent

0 2 4 6 8Quarters

D.GDP

-.5

0.5

11.

5P

erc

ent

0 2 4 6 8Quarters

D.Prices

-1-.

50

.5P

erc

ent

0 2 4 6 8Quarters

D.Policy_rate

-2-1

01

Pe

rce

nt

0 2 4 6 8Quarters

D.Gov_consumption

-1-.

50

.51

Pe

rce

nt0 2 4 6 8

Quarters

D.Priv_consumption

-4-2

02

4P

erc

ent

0 2 4 6 8Quarters

D.Investment

-4-2

02

4P

erc

ent

0 2 4 6 8Quarters

D.RER

-2-1

01

23

Pe

rce

nt

0 2 4 6 8Quarters

D.Exports

-2-1

01

23

Pe

rce

nt

0 2 4 6 8Quarters

D.Imports

- 10 -

Figure A2: Differential effects of large real shocks on first (log) differences between targeting and non-inflation targeting economies

Note: The figure shows the difference between the response of the first (log) difference of key macroeconomic variables in inflation targeting and non-inflation targeting economies to large natural disasters over the period 1970Q1-2015Q4. Statistical inference is based on 500 Monte-Carlo draws.

-.5

0.5

1P

erc

ent

0 2 4 6 8Quarters

D.GDP

-1.5

-1-.

50

.5P

erc

ent

0 2 4 6 8Quarters

D.Prices

-.5

0.5

1P

erc

ent

0 2 4 6 8Quarters

D.Policy_rate

-10

12

Per

cent

0 2 4 6 8Quarters

D.Gov_consumption

-.5

0.5

1P

erce

nt0 2 4 6 8

Quarters

D.Priv_consumption

-2-1

01

23

Per

cent

0 2 4 6 8Quarters

D.Investment

-4-2

02

4P

erc

ent

0 2 4 6 8Quarters

D.RER

-3-2

-10

12

Pe

rce

nt

0 2 4 6 8Quarters

D.Exports

-3-2

-10

12

Pe

rce

nt

0 2 4 6 8Quarters

D.Imports

- 11 -

Figure A3: Empirical density function for estimated mean output growth and inflation

Note: The figure shows the simulated density function of mean output growth and mean inflation over a horizon of four years following a large natural disaster in inflation targeting (black bars) and non-inflation targeting economies (white bars).

05

1015

De

nsity

-.3 -.2 -.1 0 .1 .2

D.GDP

01

23

4D

ens

ity

-.5 0 .5 1

D.Prices

- 12 -

Figure A4: Empirical density functions for estimated standard deviations

Note: The figure shows the simulated density function of the standard deviations of selected macroeconomic variables over a horizon of four years following a large natural disaster in inflation targeting (black bars) and non-inflation targeting economies (white bars).

05

1015

20D

ens

ity

.1 .2 .3 .4 .5

D.GDP

02

46

810

De

nsity

0 .2 .4 .6 .8

D.Prices

010

2030

De

nsity

0 .1 .2 .3

D.Policy_rate

02

46

8D

ens

ity

.2 .4 .6 .8 1 1.2

D.Gov_consumption

05

1015

20D

ens

ity0 .2 .4 .6

D.Priv_consumption

01

23

45

De

nsity

0 .5 1 1.5 2

D.Investment

01

23

4D

ens

ity

0 .5 1 1.5 2

D.RER0

24

6D

ens

ity

0 .5 1 1.5

D.Exports

02

46

De

nsity

0 .5 1 1.5

D.Imports

- 13 -

Figure A5: Response of monetary policy and interest rates to large natural disasters in targeting and non-inflation targeting economies

Note: The figure shows the response of the level (cumulated first difference) of interest rate variables in both targeting (dark shaded area) and non-inflation targeting economies(light shaded area) to large natural disasters over the period 1970Q1-2015Q4. Statistical inference is based on 500 Monte-Carlo draws.

-3-2

-10

1P

erce

nt

0 2 4 6 8 10 12 14Quarters

Policy_rate

-4-2

02

Per

cen

t

0 2 4 6 8 10 12 14Quarters

Tbill_rate

-4-3

-2-1

01

Per

cen

t

0 2 4 6 8 10 12 14Quarters

Money_rate

-2-1

01

Per

cent

0 2 4 6 8 10 12 14Quarters

Lending_rate

-2-1

01

2P

erce

nt

0 2 4 6 8 10 12 14Quarters

Long_rate

-15

-10

-50

5P

erce

nt

0 2 4 6 8 10 12 14Quarters

Cumul_policy_rate

-3-2

-10

1P

erce

nt

0 2 4 6 8 10 12 14Quarters

Real_rate

-20

-15

-10

-50

5P

erce

nt

0 2 4 6 8 10 12 14Quarters

Cumul_real_rate

-50

51

01

5P

erce

nt

0 2 4 6 8 10 12 14Quarters

Prices

- 14 -

Figure A6: Inflation and targets in euro area countries

[Continued on next page]

01

23

4

2000q1 2005q1 2010q1 2015q1Quarter

AUT

-20

24

6

2000q1 2005q1 2010q1 2015q1Quarter

BEL

-20

24

6

2008q1 2010q1 2012q1 2014q1 2016q1Quarter

CYP

01

23

2000q1 2005q1 2010q1 2015q1Quarter

DEU

-20

24

6

1995q1 2000q1 2005q1 2010q1 2015q1Quarter

ESP

-20

24

6

2011q1 2012q1 2013q1 2014q1 2015q1 2016q1Quarter

EST

-20

24

6

1995q1 2000q1 2005q1 2010q1 2015q1Quarter

FIN

-10

12

3

2000q1 2005q1 2010q1 2015q1Quarter

FRA

-20

24

6

2000q1 2005q1 2010q1 2015q1Quarter

GRC

-50

510

2000q1 2005q1 2010q1 2015q1Quarter

IRL

01

23

4

2000q1 2005q1 2010q1 2015q1Quarter

ITA

-2-1

01

23

2015q1 2015q2 2015q3 2015q4 2016q1Quarter

LTU

- 15 -

[Figure A6: Inflation and targets in euro area countries] continued

Notes: The plots show the CPI inflation rate in blue jointly with the inflation target rate (solid line) and the target corridor (dotted line) for all inflation targeting countries in the sample that are part of the euro area in 2017. The black color indicates actual data obtained from central bank websites, while the grey color indicates a hypothetical target corridor. We assume a symmetric corridor with a size of +/- 1 percentage points around the target rate.

Country codes: AUT=Austria, BEL=Belgium, CYP=Cyprus, DEU=Germany, ESP=Spain, EST=Estonia, FIN=Finland, FRA=France, GRC=Greece, IRL=Ireland, ITA=Italy, LTU=Lithuania, LUX=Luxemburg, LVA=Latvia, MLT=Malta, NDL=Netherlands, PRT=Portugal, SVK=Slovakia, SVN=Slovenia..

Figure A7: Inflation and targets in non-euro area countries

[Continued on next page]

01

23

4

2000q1 2005q1 2010q1 2015q1Quarter

LUX

01

23

2014q1 2014q3 2015q1 2015q3 2016q1Quarter

LVA

01

23

45

2008q1 2010q1 2012q1 2014q1 2016q1Quarter

MLT

01

23

4

2000q1 2005q1 2010q1 2015q1Quarter

NLD

-20

24

6

2000q1 2005q1 2010q1 2015q1Quarter

PRT

02

46

2005q3 2008q1 2010q3 2013q1 2015q3Quarter

SVK

02

46

8

2007q3 2009q3 2011q3 2013q3 2015q3Quarter

SVN

12

34

5

2009q3 2011q1 2012q3 2014q1 2015q3Quarter

ALB

02

46

1995q1 2000q1 2005q1 2010q1 2015q1Quarter

AUS

05

1015

20

2000q1 2005q1 2010q1 2015q1Quarter

BRA

- 16 -

[Figure A7: Inflation and targets in non-euro area countries] continued

[Continued on next page]

-20

24

6

1990q1 1995q1 2000q1 2005q1 2010q1 2015q1Quarter

CAN

-50

510

2000q1 2005q1 2010q1 2015q1Quarter

CHL

05

1015

2000q1 2005q1 2010q1 2015q1Quarter

COL

05

1015

1997q3 2002q1 2006q3 2011q1 2015q3Quarter

CZE

01

23

45

1990q1 1995q1 2000q1 2005q1 2010q1 2015q1Quarter

GBR

510

1520

2007q3 2009q3 2011q3 2013q3 2015q3Quarter

GHA

23

45

67

2010q1 2011q3 2013q1 2014q3 2016q1Quarter

GTM

05

10

2000q1 2005q1 2010q1 2015q1Quarter

HUN

05

1015

20

2005q3 2008q1 2010q3 2013q1 2015q3Quarter

IDN

05

1015

20

2000q1 2005q1 2010q1 2015q1Quarter

ISL

-50

510

1997q3 2002q1 2006q3 2011q1 2015q3Quarter

ISR

02

46

2000q1 2005q1 2010q1 2015q1Quarter

KOR

- 17 -

[Figure A7: Inflation and targets in non-euro area countries] continued

Notes: The plots show the CPI inflation rate in blue jointly with the inflation target rate (solid line) and the target corridor (dotted line) for all inflation targeting countries (non-euro area) in the sample. The black color indicates actual data obtained from central bank websites, while the grey color indicates a hypothetical target corridor. We assume a symmetric corridor with a size of +/- 1 percentage points around the target rate.

Country codes: ALB=Albania, AUS=Australia, BRA=Brazil, CAN=Canada, CHL=Chile, COL=Colombia, CZE=Czech Republic, GBR=United Kingdom, GHA=Ghana, GTM=Guatemala, HUN=Hungary, IDN=Indonesia, ISL=Iceland, ISR=Israel, KOR=Korea, MEX=Mexico, NOR=Norway, NZL=New Zealand, PER=Peru, PHL=Philippines, POL=Poland, ROU=Romania, SRB=Serbia, SWE=Sweden, THA=Thailand, TUR=Turkey, ZAF=South Africa.

24

68

2000q1 2005q1 2010q1 2015q1Quarter

MEX

-20

24

6

2000q1 2005q1 2010q1 2015q1Quarter

NOR

02

46

8

1990q11995q12000q12005q12010q12015q1Quarter

NZL

-20

24

6

2002q12005q3 2009q12012q3 2016q1Quarter

PER

02

46

810

2002q1 2005q32009q12012q3 2016q1Quarter

PHL

05

10

2000q1 2005q1 2010q1 2015q1Quarter

POL

-20

24

68

2005q32008q12010q32013q12015q3Quarter

ROU

05

1015

2009q32011q12012q32014q12015q3Quarter

SRB

-20

24

1995q12000q12005q12010q12015q1Quarter

SWE

-20

24

68

2000q1 2005q1 2010q1 2015q1Quarter

THA

24

68

1012

2006q1 2008q32011q12013q3 2016q1Quarter

TUR

05

1015

2003q32006q32009q32012q32015q3Quarter

ZAF